Changelog
Complete release history and changes for Beyond Better
Changelog
All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning.
[0.14.3] - 2026-11-18
IMPORTANT: Major New Features
π Personas
Set BB up once for each kind of work and get consistent results without
repeating yourself. A persona is a named setup of model, guidelines, tool sets
and limits for the orchestrator, agent or admin role. Personas replace the
per-role model settings, and are managed from a new Personas tab in
workspace and project settings.
What’s new:
- Built-in default personas, plus your own, set per workspace and project
- Choose a persona for each role in chat options, and “Customise for this
conversation” to change its model settings for one conversation - Limits on how deep agents may delegate, how many run at once, which personas
they may start, and approval before an agent goes outside its task - A simpler Edit Persona dialog: the essentials up front, the rest under
Advanced - A conversation keeps the persona it started with when defaults change
- Existing role model settings become personas automatically
π Background agents
Hand BB bigger jobs and keep working while they get done. The orchestrator now
starts agents in the background and carries on, and each agent reports back
when it finishes.
What’s new:
- Up to 5 agents run at once by default, and agents can start agents of their
own, up to three levels deep, when their persona allows it - The orchestrator can message, cancel, wait for and read the history of its
running agents - Agents and the orchestrator can ask a question and wait for the answer from
anyone who can write to the collaboration - Agent groups show each agent’s task, persona, status, elapsed time and latest
progress, with Message and Cancel buttons - The status bar says what each running agent is doing
- The delegate_tasks tool is renamed start_agents
π Agent Accounts
Team background work no longer has to depend on one person’s access. Agents in
shared team workspaces can run as workspace service accounts instead of as the
person who started them.
What’s new:
- Agent Accounts managed in workspace settings, bound to personas and projects,
with a policy set at organization, workspace or project level - A “Run my agents as me” switch in the chat toolbar
- Agent groups show which account a run used and who started it
- Usage analytics by account, with an agent account’s usage split by who
started its runs
π Web App on phones and tablets
Keep up with your work away from your desk. The Web App now works on phones
and tablets, and can be installed as an app.
What’s new:
- A top bar and navigation drawer on phones, a compact side rail on tablets,
and touch-sized controls throughout - Install from the browser: Add to Home Screen on iPhone and iPad, Install on
Android and desktop - Chat on phones: info panels and the message filter open as sheets, the
message box stays above the keyboard, and wide tables and code scroll within
their message - Settings on phones: a list of sections, each opening a full-width page, with
save buttons in reach - Project and skill folders browsed one folder at a time
- BB Bridges, Doctor and the guided tour still need a larger screen
Added
- Web App on phones and tablets: a top bar and navigation drawer on phones, a compact side rail on tablets, and touch-sized controls throughout
- install the Web App as an app from the browser (Add to Home Screen on iPhone and iPad, Install on Android and desktop)
- chat on phones: info panels and the message filter open as sheets, the message box stays above the keyboard, and wide tables and code scroll within their message
- settings on phones: a list of sections, each opening a full-width page, with forms that fit the screen and save buttons that stay in reach
- simpler Hub Connection page on phones, with a link to connect to another Hub address
- browsing project and skill folders on phones, one folder at a time
- personas: named setups of model, guidelines, tool sets and delegation for the orchestrator, agent and admin roles, with built-in defaults, set per workspace and project
- persona per role in chat options, with “Customise for this conversation” to change its model settings for one conversation
- persona limits: how deep agents may delegate, how many run at once, which personas they may start, and approval before an agent uses tools or data sources outside its task
- background agents: the orchestrator starts agents and carries on, and each agent’s report arrives when it finishes; up to 5 agents run at once by default
- the orchestrator can message, cancel, wait for and read the history of its running agents
- agents can start agents of their own, up to three levels deep, when their persona allows it
- the orchestrator and agents can ask a question and wait for the answer; anyone who can write to the collaboration can answer
- agent groups show each agent’s task, persona, status, elapsed time and latest progress, with Message and Cancel buttons for each agent
- status bar says what the current tool is doing, e.g. “Editing src/foo.ts”, with a line for each running agent
- Agent Accounts: workspace service accounts that agents run as, managed in workspace settings and bound to personas and projects, with a policy at organization, workspace or project level
- “Run my agents as me” switch in the chat toolbar; agent groups show which account a run used and who started it
- usage analytics by account, with an agent account’s usage split by who started its runs
- multi-user collaborations show everyone’s queued messages and whose message is being answered; remove your own queued message, and Stop offers dropping your message, halting all or cancelling an agent
- each person’s messages, icon and answers get their own colour in multi-user collaborations
- sign-in prompts for MCP servers that need each user to sign in: listed above the chat input, shown in the conversation when a tool call waits for sign-in, and on failed MCP tool calls
- Workspace Guidelines written directly in workspace settings, used by every project in the workspace
- delete an organization from Settings > Organization > General, straight away or at the end of the billing period, with its data purged 30 days later
- page for signed-in people without an organization, listing pending invitations and offering a personal organization
- Alerts tab where organization owners and admins set their own credit alert amounts
- email to organization owners when auto top-up can’t run because no payment method is set up
- Bedrock cross-region inference profiles (us., eu., global. and others) in the model picker
- model pickers flag deprecated models with their retirement date and replacement
- upload a Google service account key file in organization settings
- a project can stop using a tool key set for its workspace, e.g. the web search key
- confirmation before resetting an AI provider to its defaults
Changed
- refreshed Web App look: consistent buttons, colours, icons and dialogs in light and dark mode
- settings tabs on phones are a row that scrolls sideways, replacing the drop-down list
- BB Bridges, Doctor and the guided tour are left out of the phone layout; use a larger screen for them
- Breaking Change: self-hosted Hubs with Google service account keys look up each key by organization; rename stored key secrets to
org-<organizationId>/<name>, or setapi.allowLegacyGoogleSaSecretNames: trueuntil they are renamed. Not needed for single-user Hubs or a Hub limited to one organization. - BB (Hub, Web App, CLI, Desktop App and BB Bridge) is now licensed under the Beyond Better End User License Agreement instead of AGPL-3.0; earlier releases remain under AGPL
- Personas tab replaces the Models tab in workspace and project settings; existing role model settings become personas automatically
- a conversation keeps the persona it started with when workspace or project defaults change
- renamed the delegate_tasks tool to start_agents
- notice in chat options when a model change may use more credits
- simpler Edit Persona dialog: name, model and tool sets up front, with model parameters, guideline files, limits and report style under Advanced
- Agent Accounts tabs shown in personal workspaces, explaining that agents there always run as you
- workspace and project guidelines files both load; a project’s file no longer replaces the workspace’s
- reorganized settings: Organization in seven tabs (General, People, Workspaces, Connections, Billing, Usage, Alerts), workspace and account Details renamed General, MCP Servers renamed Connections, Archives renamed Export & Import
- API keys, data source credentials, MCP server secrets and sign-in tokens, and tool keys are kept in an encrypted secret store and only ever shown masked, with password fields for keys in tool config editors
- only team leads, owners and organization admins add or change a workspace’s MCP servers and workspace skills; members still sign in to MCP servers themselves
- organization owners and admins can see and manage API tokens and MCP servers of every shared workspace
- saved selections of retired models use their replacement, with a notice in the conversation; retired Claude models point at Anthropic’s current replacements, and Claude Sonnet 4.5 is retired
- Claude Haiku 4.5 charged at its published prices
- credit alert amounts follow the plan and auto top-up minimum, with no “top-up coming” alert while auto top-up is paused or at its daily limit
- auto top-up minimum is $5.00 when not set and at least $1.00; a declined top-up pauses retries for a while
- context clearing starts at a share of each model’s context window, so models with large windows keep loaded files and memories longer
- after signing in, BB returns to the last-used organization, else the personal one
- messages typed while BB restarts are held and sent when it is back; queued messages and background agents’ reports survive a Hub restart
- BB Bridge app named “BB Bridge” on macOS, Windows and Linux, with a drag-to-Applications macOS disk image
- clearer error messages from the Hub, shown as they are in the Web App, and MCP failures described by kind
- more reliable MCP sign-in, token renewal and reconnects
- improved security for model hosts, organization membership, API tokens, Google service account keys, project tool settings, logs and data access limits
Fixed
- staying signed in after confirming an email link or code
- no workspace being selected after signing in
- deep links opening a project in the wrong workspace
- the first chat message on a new Free plan raising the low-balance alert
- auto top-up never running when its minimum was not set
- a plan upgrade fully covered by a coupon, or to a free plan, showing as pending until the page was reloaded
- images and PDFs whose file type didn’t match their content breaking the conversation
- conversations stuck on message-history errors the model provider rejects
- Bedrock inference profile IDs being refused or sent unsupported settings
- Ollama model picker search, tools-only filter and paging doing nothing
- remote Ollama hosts listing every model instead of the chosen ones
- a saved Ollama Cloud key being sent to a local Ollama host
- AI provider Reset not removing the organization’s settings and saved key
- discovered models disappearing from model pickers when a provider’s discovery was skipped
- MCP sign-in and connect reporting success when they failed, and an authorised MCP server’s tools silently missing from the workspace’s tool sets
- failed project, workspace and data source settings saves reported as saved
- teammates seeing stale settings after another member changed them
- browser Back and Forward not changing the settings tab
- settings and onboarding tours that never started or pointed at missing elements
- restoring a project archive dropping saved keys the archive had no value for
[0.14.2] - 2026-11-04
Added
- public demo: try BB without an account, with a sample project, a sample team and guided examples; everything made is kept on signing up
- BB Files data source: files stored by BB for a project, with no external account or software
- new projects start with a ready BB Files data source, a README and guided examples for a first chat
- multi-user collaborations: several people can follow and add to the same collaboration, with each message showing who wrote it
- organization Members tab: list members, change roles, remove members and leave an organization
- organization Teams tab: create, rename and delete shared teams, manage team members and create a team’s workspace
- organization invitations by email, sent and resent from the Members and Teams tabs, with pending invitations listed
- transfer of organization ownership to another member
- Workspaces tab in organization settings, where owners and admins manage every shared workspace
- separate billing contact email for invoices and receipts
- organization credit pool, with team balances on Plans & Credits and credit transfers between teams and the pool
- choice of the team that receives purchased, coupon, upgrade and signup credits
- Advanced model config section for temperature in model settings and chat options
bb chatanswers tool approvals and form requests in the terminal
Changed
- Breaking Change: local and self-hosted BB (hub/API) versions older than 0.14.0 can no longer save or read data source connections with BB Cloud, and must be upgraded. Existing connections are kept and appear after upgrading.
- MCP servers are no longer included in the Free plan; existing MCP server settings are kept but their tools no longer load
- Archives export and import are unavailable for now while they get more testing
- unused credits expire one year after they are granted
- usage draws on the team’s credit first, then the organization credit pool
- a deleted team’s remaining credit moves to the organization credit pool or a chosen team
- auto top-up credits the organization credit pool, and its daily limit applies to the whole organization and resets at midnight UTC
- cancelling a subscription always takes effect at the end of the billing period
- an organization moving to the Free plan becomes personal when the downgrade takes effect, not when it is scheduled
- BB Bridges are placed in teams, and team members can grant their own BB Bridge to their team’s workspaces
- team leads can make changes in their team’s workspaces
- API tokens reach every workspace of the teams their service account belongs to
- view-only users get a read-only collaboration, with chat input, feedback answers and Stop disabled
- new, renamed, starred and deleted collaborations show up straight away for everyone with the project open
- only project owners can delete or restore collaborations
- clearer handling of deleted workspaces and teams, with controls disabled and a plain explanation
- clearer error messages across the Web App, CLI and Hub
- improved security for web page fetching, sign-in, roles and permissions, API tokens, billing and organization data
- removed Homebrew packaging
Fixed
- BB Bridge: editing a file it cannot read as text no longer overwrites it.
- BB Bridge: file, git and command errors are shown in plain words.
- subscription renewals and scheduled plan changes not being processed
bb chatnot connecting to collaborations- view-only users unable to open some collaborations, and opening one adding an empty conversation each time
- auto top-up charging an organization with several teams more than its daily limit
- amounts under $1 in usage alerts missing their leading zero
- Unsubscribe link shown in account emails
- plan change Confirm sending more than one request
- billing details and access briefly showing the wrong state while loading
Security
- BB Bridge: please update. Earlier bridges could be made to write outside a shared folder
through a symbolic link inside it, or accept a shared folder whose path only looked like it was
inside an allowed one. Paths are now resolved before they are checked. - BB Bridge: git commits stay inside the shared folder. A shared folder inside a larger git
repository, or linked to a repository elsewhere, no longer auto-commits, and nothing under.git
can be changed through the bridge. - BB Bridge: creating a file with an edit now needs write permission as well as edit permission.
[0.14.1] - 2026-10-31
IMPORTANT: Major New Features
π One Reasoning effort setting for every model
How much a model thinks is now set with a single Reasoning effort level,
replacing the separate thinking toggle, budget and effort controls, and it works
across all model providers.
What’s new:
- One Reasoning effort select (auto, off, low up to max) in role model settings,
project settings and chat options, offering only the levels each model supports - Supported for Anthropic, Bedrock, OpenAI, Grok, Gemini, DeepSeek, OpenRouter and
Ollama models, including extended thinking for Claude models on Bedrock - The model info panel shows the reasoning used by the last request
- Temperature follows each model’s recommended default and is no longer shown in
model settings - Existing thinking settings are converted automatically
π Prompts tuned for Claude 5 and Claude Fable
BB’s system prompts, tool descriptions and data source instructions have been
rewritten for current models.
What’s new:
- Dedicated system prompts for Claude 5 models and for Claude Fable 5 and 5.1,
for both the main conversation and delegated task agents - Tool descriptions and data source instructions rewritten to be clear and
direct, for every model - Prompts no longer ask models to write out their reasoning, which current
models can decline - Progress updates from Claude 5 models appear in the conversation, even when
they arrive as thinking only
π Flexible archives
Archives can now be imported almost anywhere, and are managed from a new
Settings > Archives page.
What’s new:
- Export a workspace, several projects or chosen collaborations, from the
Web App or CLI - Import into an existing workspace or project, not only the one the archive
came from - When items already exist, stop the import or keep them and import the rest;
deleted projects and collaborations are restored - Data source credentials can be included, encrypted with a passphrase; MCP
servers and skills come along too - An import report shows what was imported, restored, kept or failed, and why
Added
- brute-force protection for sign-in, with captcha after repeated failed attempts and temporary lockout
- request limits for signup, password reset and email verification
- sign-in security monitoring for self-hosted Hubs, with a read-only monitoring token
- setting for the client IP header when the Hub runs behind a proxy
- Settings > Archives page for exports and imports, with progress and results
- export and import of collaborations and of several projects at once, in the Web App and CLI
- MCP servers, skills and passphrase-encrypted data source credentials in archives
- choice to stop or keep existing items when an archive import finds them
- import report showing imported, restored, kept and failed items, with reasons
- single Reasoning effort setting for all model providers
- system prompts tailored for Claude 5 and Claude Fable models
- sign out of a single device; idle devices signed out automatically
- failed model requests and tool results shown in the collaboration log
- Skills button in collaboration header toolbar
Changed
- provider error details logged when an OpenRouter request fails
- archives can be imported into an existing workspace or project, not only the one they came from
- deleted projects and collaborations restored when importing an archive
- archives from older versions upgraded on import; archives from newer versions refused
- archives and storage migrations include all collaboration records and uploads
- CLI sign-in explains rate limits and captcha
- rewrote system prompts, tool descriptions and data source instructions for current models
- temperature follows each model’s recommended default; removed from model settings
- data source access settings apply to every operation, including LLM tools
- improved security for data sources, skills, archives and Hub maintenance features
- improved search and faster file details for BB Bridge, Google Drive and MCP data sources
- better handling of MCP server changes, restarts and connections
- more reliable multi_model_query and delegated tasks
- models understand resource references added from chat input
- clearer messages when a data source can’t rename, move or remove a resource
- more robust self-update, with rollback if an install fails
- storage backend migration limited to moving between local and cloud
- plan upgrade credits are the difference in signup credits between plans
- removed Export Settings buttons, which did nothing
- billing changes (plan, payment methods, credit purchases, auto top-up, coupons, billing address) limited to organization owners and admins; other members can still view billing, with the controls disabled
- plan upgrades charge the card before the plan changes; a declined card leaves the current plan in place
- web search usage in BB cloud is charged to credits
- a successful retry of a failed renewal payment restores the plan, and Stripe’s retry settings decide when an unpaid plan ends
- improved security for billing, usage and account data
- plan stays active while a failed renewal payment is retried, with a notice to update the card; cancelling during that time ends the plan straight away without taking the overdue payment
- deleting a user keeps their organization’s usage and billing history
- switching organization opens its last-used workspace
- a scheduled downgrade can be replaced with another, or withdrawn with “Keep current plan”
- plan upgrades of under 50 cents are not charged
- plans upgraded with a time-limited 100% coupon are billed by Stripe when the coupon ends
- storage migration reports failure when files could not be copied, and the old storage can’t be destroyed while it holds files the new one is missing
Fixed
- Google Doc and Sheet names showing URL encoding and losing hyphens when created
- failed statements returning success (“Error handling statement”) instead of an error
- images loaded into a conversation breaking every later OpenRouter request
- Hub admin endpoints rejecting requests that use only the admin token
- switching a conversation between model providers, e.g. from OpenRouter or Ollama to Claude
- requests rejected for temperature or thinking settings on Gemini, OpenAI and Bedrock models
- Claude 5 models still thinking, and using thinking tokens, with reasoning turned off
- chat options carried over between collaborations
- Google data source tokens no longer refreshing after an organization adds its own OAuth client
- changing a memory’s scope losing its details or overwriting another memory
- exports, migrations and archive imports missing records
- destroying one storage backend removing data from the other
- Doctor reporting errors for a correct setup
- Desktop App not reopening after a macOS update
- CLI commands not exiting after finishing
- cancelling a plan and adding a card from Plans & Credits
- billing, organization settings and workspace details showing or saving the previous organization’s or workspace’s data after switching
- requests using the previous organization’s team or workspace after switching organization
- errors instead of “Select a workspace first.” when no workspace is selected
- moving storage to the cloud or importing an archive reporting files copied before their upload had finished or after it failed
- auto top-up failing when an organization had more than one default card
- plan downgrades not taking effect at the end of the billing period
- organizations left without a plan when a paid subscription ended
- a pending downgrade still applying after a later upgrade or cancellation
- the previous plan still being billed after upgrading with a coupon that covers the full price
- team and organization usage analytics missing former members’ usage or counting usage from other teams
- model usage occasionally charged twice
- model and web search usage billed to another workspace after switching organization or project mid-conversation
- subscription coupons showing “Invalid coupon code” instead of the reason they can’t be used
- a double-clicked plan upgrade charging twice
- replacing or removing a saved card, and showing why a removal is refused
- cards saved during payment setup not being stored
[0.14.0] - 2026-10-26
Added
- visibility toggle for password fields
- UI control for request feedback for tool calls
- GuidelinesRefreshMode to control frequency of reading guidelines file
- tracking and metadata for guidelines and system prompt for each interaction
- ingestion submit endpoint includes collaborationConfig.guidelinesRefresh
- support move_resources tool for google data source
- support for creating directories for google data source
Changed
- dedicated table for data source config
- updated Google auth handling for data sources
- updated BB Bridge to use data source connection primary key
- more fine-grained control of per-tool request feedback settings
- removed deprecated keys from global config
- bumped config.yaml version to 2.5.0
- renamed workspace/project
configAPI endpoints tosettings - removed deprecated config keys from CLI
- removed deprecated config keys from DUI
- added token endpoint auth method metadata to MCP OAuth types
- removed metadata timestamp from google and notion markdown conversion
- reduced read/write cycles for system prompt contents
- include reasoning arg for openrouter provider
- better handling of transient supabase network failure for llm-proxy
Fixed
- improved handling of data source grants
- improved handling of BB Bridge grants
- honour BB_GLOBAL_CONFIG_FILE for all config file reads
- min/max/step values in BUI for toolResultsLimits
- cache token usage for openrouter provider
- destroying user sessions after transient supabase network failure
- using wrong SA credentials after transient supabase network failure
- tracking all connection types for drain manager
- removing items from lists in workspace and project settings
- custom tool sets shown with builtin tool sets
- display of tool set cards in BUI
[0.13.7] - 2026-10-20
Added
- show hub connection details in login and home pages of bui
Changed
- split config for buiBaseUrl and oauthCallbackUrl
- updated handling and templates for auth flows
Fixed
- callback URL handling for oauth
[0.13.6] - 2026-10-17
Added
- support in DUI to disable running BUI
- support to generate TLS certs in DUI
- enforce max size of tokens for tool results
- improved error reporting when LLM token max is exceeded
Changed
- handling of API responses for Google data source
- added better chunk handling for malformed apply_path input from LLM
- loading ranges and partial data for Google Sheets
- bounce/retries for MCP connection failures
- http protocol fallback when making API requests from BUI
Fixed
- version checks for updates in DUI
- cookie reset for Google file picker in DUI
- OAuth redirect URL for Google
- handling of context management and effort for latest Claude models
- more robust deployment processes
- Slider UI in Firefox
- reading default scopes for MCP server OAuth flow
[0.13.5] - 2026-10-13
Added
- support for Claude Haiku 5.5
- show current hub under status in side bar
Changed
- hide Google oAuth access token in BUI
- redact internal data from status endpoints for cloud API & BUI
- loading of app settings once with resilient fallbacks
Fixed
- deleting filesystem datasources when editing in cloud BUI
- overwriting Google access tokens when editing datasource
- oAuth flow failing for local self-hosted API
- loading internal uploads datasource in cloud API (regression)
[0.13.4] - 2026-10-07
Added
Changed
- uploaded files using new contentstore datasource
Fixed
- show informative error when billing address is missing during payment flow
- linking Stripe customer to organization rather than user
- data serialization during logging for self-hosted API
- uploading files on cloud API
- failing cross-user writing to content store for fallback files
[0.13.3] - 2026-09-30
Added
- added BB as configurable LLM provider
- custom remote browser service for fetch tools
- support for GPT Astra 6
- support for Claude Opus and Sonnet 5.5
Changed
- model classes for new openai models
Fixed
- bare version reported from api and bui commands
- global error catching in BUI
- permissions for chat window in app (DUI)
[0.13.2] - 2026-09-03
Added
Changed
Fixed
- break hub reset loop for invalid user auth
- release artifacts for BB Bridge
[0.13.1] - 2026-09-03
Added
- editing name/description for workspaces and organizations
- editing name for user accounts
- support for AWS Bedrock
Changed
- UI cleanup in BB Bridge
- UI cleanup in BUI
- support for user-selectable capabilities in data sources
- renamed orgType from company to shared
- automatically change org to shared following plan upgrade
- LLM providers config supports multiple auth methods
Fixed
- BB Bridge (LWB) included in build artifats
- stale Google oAuth creds after re-auth
[0.13.0] - 2026-08-27
Added
- new subscription plans
- usage quotas for orgs
- PDF file parsing for OpenRouter provider
Changed
- code refactor for large files
- removed support for legacy migrations
- load resources uses faster write-behind for bucket storage
- use per-role model configs for workspace and project settings
Fixed
- triggers for token counts for context compaction
- formatting for MCP tools
- confusion with related vs parent interaction ID for agent tasks
- more robust dynamic model discovery
- confusion with user RBAC and subscription plan features
- missing captcha for signup form
[0.12.1] - 2026-08-16
Added
- support for ChatGPT 5.6
- support for DeepSeek 4
- support for Grok 4.6
- organization scoped dynamic model discovery
- support OpenRouter as LLM provider
- model selection for for providers with dynamic model discovery
- copy model ID from model system card
- support caching and context compaction for Grok models
- browsable API docs (using Swagger UI)
Changed
- LWB uses Happy Eyeballs dual-stack connection racing
- added support for additionalPaths to LWB
- switch to Responses API for ChatGPT models
- switch to Responses API for Grok models
- pooled instances for fetchManager (webpages and screenshots)
- explicit provider for model ID
- organization scoped model registry
- updated ingestion endpoint to accept collaborationConfig
- report config validation errors for data sources
Fixed
- ensure hydrated Supabase session for service accounts
- auth calls for LWB pairing
- safe updates for LWB connections
- LWB auth checks against agent user
- avoid onboarding for ‘special’ user accounts
- database string collation
- high CPU load in LWB
- exposure of provider API keys in browser
- Ollama Cloud model discovery
- including background and resources for delegated agent tasks
- sticky sessions for prompt caching for OpenAI models
- grouping of auxiliary log entries for agent tasks
- separation of parent vs related log entries
- model selection for delegated agent tasks
- auth error handling for llm-proxy requests
- prevent fatal error when config validation fails for data sources
[0.12.0] - 2026-08-04
IMPORTANT: Major New Features
π Introducing BB Bridge
Work with BB from anywhere β cloud or local β while keeping full access to the
files and tools on your own workstation.
BB Bridge is a lightweight agent you install on your computer that connects it
securely to a BB Hub (cloud or self-hosted). Once paired, you can create a
data source backed by your workstation from any project, without needing to run
the full BB API locally.
What’s new:
- New BB Bridges settings tab for pairing, naming, and managing your devices
- Install the BB Bridge app, pair it with your Hub, and it’s ready to use
- Add a BB Bridge data source to any project and select the paired agent β
same file access, search, and command execution you get from a local filesystem
datasource, just reachable from anywhere - One bridge can serve multiple data sources/projects at once
Note: access to a bridge is scoped to what you explicitly allow (root
folder, allowed commands) β the bridge owner is always in control of what’s
exposed.
π Introducing a new trust & authorization system
Replaces the old localMode setting with a proper trust model that scales from
fully offline use to shared team workspaces β and lays the groundwork for
stronger security as BB grows into more multi-agent and multi-user scenarios.
What’s changing:
localModeis retired in favor of an explicit trust system that adapts to
how BB is deployed β offline, self-hosted, or cloud- Establishes verified trust between BB Bridge and its Hub, so a paired device
only ever does what it’s been explicitly authorized to do - Foundational work for cryptographic identity β device pairing now sets up
the trust relationship that future features (fine-grained permissions, audit
logging, per-request authorization) will build on
Why it matters: this is the plumbing that makes BB Bridge safe to use across
teams β every request a bridge handles will be traceable to real, verifiable
authorization, not just an open connection.
Added
- support for Claude Opus 5
- support for Gemini Flash 3.6
- support for Gemini Flash Lite 3.5
- trustBoundary config
- dataResidency config
- testing for delegate_tasks tool
- testing for orchestrator and agent controllers
Changed
- moving memory between scopes
- localMode changed to authMode
- runtimeEnvironment change to runsIn
- check valid project and datasource before saving project during auto-provisioning
- updated preset values for context compaction trigger
- search all datasource async for suggest resources
- more robust handling of invalid LLM responses
- more robust handling of Cloudflare errors
- improved git commit messages
- confirmation dialog when removing data source from project
Fixed
- collaborationId when saving memories
- readOnly support for project provisioning templates
- crash when displaying message entry with invalid markdown formatting
- token usage recording for Google LLM
[0.11.25] - 2026-07-16
Added
Changed
Fixed
- Added callback for MCP server token refresh
[0.11.24] - 2026-07-15
Added
Changed
- allow request context resolution from projectId
Fixed
- MCP token refresh using single-flight requests
[0.11.23] - 2026-07-12
Added
- support for Claude Sonnet 5
- support for Claude Fable 5
- support for Ollama cloud
- configure LLM providers in Organization Settings
- support for injecting “correction” prompt into actively running statement
- generate images, including from collaboration using non-image models (using generate_artifact tool)
- storage and management for generated artifacts (images)
- send email when credits getting low
Changed
- show coupon credit purchase in invoice history
- cancelling actively running statement improvements (dynamic state for send button)
- encourage LLM to read pinned memories
- moved GST handling from internal billing to Stripe
- show user feedback when Fable refuses to process request
- dynamic management of provider models
- input/output modalities for models
- write_resource tool can save artifacts to data source
- layout tweak for chat input
- return collaboration URL in responses from chat API endpoint
- changed location of resource revisions from database to content storage
Fixed
- removed resources committed to git
- max slug length for volume names in deno sandbox
- saving auto-topup settings
- handling of resource not-found errors in datasources
- focus of input editor when switching markdown and visual modes
- pasting markdown content into visual editor
- billing charges for cached tokens
- saving metadata for memories
- added missing values for model capabilities
- validation of tool input schema (fix error with gemini requests)
- Vary origin for API requests to fix cache-control CORS errors
[0.11.22] - 2026-06-28
IMPORTANT: Major New Features
π Introducing Skills
Beyond Better now supports Skills β reusable, installable instruction packages that teach BB how to handle specific tasks. Each Skill bundles a set of instructions with optional scripts, reference files, and assets, scoped to a project or your whole workspace.
Install Skills from any public URL (including GitHub repositories and community skill libraries), or use the file browser in BB’s interface to select a local Skill from a markdown or zip file β uploading your own Skill files is also supported. Once installed, Skills live in your library and can be activated on demand mid-conversation, giving BB specialised knowledge exactly when you need it.
Skills are available to all users. Anyone can create and share them, making it easy to encode team workflows and hand them across projects. Browse what’s available β or build your own.
π Introducing Sandboxes with Deno Deploy
Beyond Better now supports Sandboxes β an ephemeral execution environment powered by Deno Deploy that BB can use as a datasource. The Sandbox gives BB a writable workspace filesystem and command execution capability: stage files, run scripts, test code, and transfer results across your other datasources, all within a single conversation.
Skills and Sandboxes are designed to work together. A Skill that includes executable scripts can use the Sandbox as its runtime β unlocking data transformation, code validation, and private API integrations. Enterprise teams can use this combination to build a self-hosted BB API hub.
This release requires your own Deno Deploy account and API key β BB-managed Sandboxes are on the roadmap. There are no tier restrictions; all users can connect a Sandbox today. Set up your account at deno.com/deploy and add your API key in BB Settings.
Added
- support for Skills
- support for sandbox using Deno Deploy
- diagnostics support for corrupt LLM responses
Changed
- allow STDIO for single-tenant (enterprise) cloud API hubs
- improved storage keys for localStorage
- distinction between extended and adaptive thinking
- deprecated
run_commandtool; addedexecutetool - GST calculated at Stripe rather than added to model price
Fixed
- handling of request context for path parts
- normalize line endings in scopes for google service accounts
- management of storage size for localStorage
- oAuth discovery endpoints for multi-tenant servers
- list display of tools for project vs workspace
- display of MCP tools
- passing interactionId when submitting ingestion request
- ignoring query params when loading collaboration page
- missing cache token pricing for anthropic models
- better recovery empty LLM responses
[0.11.21] - 2026-06-01
Added
- org restrictions for custom hubs
Changed
Fixed
[0.11.20] - 2026-05-31
Added
- pinned memories
rgincluded in default list of allowedCommands for run_command tool- large results storage for interactions to allow LLM to query rather than process datasets
- gmail queries for google data source; uses results storage for large queries
- support for Opus 4.7 and 4.8
- support for Gemini 3 Flash
Changed
- removed scopes for API tokens
- cancelling active statement for collaboration also cancels all delegated agent tasks
- data source instructions respect mode parameter
- updated to deno v2.8; lots of type fixes
- Data source config for google supports ‘gmail’ mode
Fixed
- loading service account details for google auth
- loading guidelines from google data source
- normalizing URL for non-filesystem data sources
- searching google data source with multiple shared drives
- suppor for adaptive thinking for newer claude models
[0.11.19] - 2026-05-11
Added
- [beta] RBAC permissions system (foundation for granular access control)
Changed
- ranked searching for memory files
- LLM instructions for loading memory files
- updated user onboarding process to support role assignment
- switched API tokens from scopes to role assignment
- subscription feature checks using RBAC system
Fixed
- google access for multiple workspace drives
- premature destruction of image array in resource endpoint
[0.11.18] - 2026-05-06
Added
- support data source selection for LLM Guidelines file
- searching and sorting for projects in Project Manager
Changed
- card layout of projects in Project Manager
Fixed
[0.11.17] - 2026-05-05
Added
- support for API tokens to allow external services to call BB
- automatic project provisioning, using templates
- UI for service account auth for google data source
- Ingestion endpoint for sync response
Changed
- reduced list of memories to 30 for memory tool; uses weighted sorting
- adaptive thinking effort for new Anthropic models
- use 1 hour cache for system prompt and tools
Fixed
- OpenAI tool_use_id resolved properly
- Event ingestion with API tokens
- Project listing doesn’t fail with invalid datasource auth config
[0.11.16] - 2026-04-24
Added
- Rich-text editor for prompt input (markdown editor still available)
- Display of attached resources in prompt editor
- [beta] Conversations from outside BB (email, slack, etc)
- [beta] Approvals system for conversations originating outside BB
- support for adding credit via coupons
- [beta] multiple secrets providers for auth-sensitive data sources
- [beta] support for service account (with DWD) for google data sources
- [beta] org-wide setting for data source auth
- support for chatgpt 5.4
Changed
- improved matching and results for suggest resources
- improved handling of resource suggestions across multiple data sources
- improved use of ‘delegate tasks’ tool - LLM more likely to delegate complex tasks - better reporting of agent task results
- model defaults
- support for adaptive thinking to replace extended thinking
Fixed
- improved recovery when LLM API returns invalid response
- propagating settings changes from frontend to API processes
- read cache performance and pruning
[0.11.15] - 2026-03-29
Added
- searching spreadsheets for google data source
- close button for google file picker
Changed
- setting of root folder for google data sources, using google file picker
Fixed
- field entry in modal dialogs, eg custom tool sets
- read-only restriction for google data source
- saving new documents within root folder for google data source
[0.11.14] - 2026-03-24
Added
- editing memories
Changed
- use model class for defaults
Fixed
- improved websocket reconnect handling and notifications
- loading memories for new collaborations
[0.11.13] - 2026-03-23
Added
- consistent UI elements throughout the app
- better a11y support across UI elements
Changed
- updated onboarding process to create new org/workspace as needed
- redesigned toast notifications
- improved stying and layout for chat input
Fixed
- saving workspace and project settings
- more reliable loading of current model details in chat
- various bug fixes
[0.11.12] - 2026-03-16
Added
- simplified Chat UI (removed complex headers, moved buttons to menu)
- diagnostic info for Hub selection
Changed
- improved model selection (always use latest version for model family and class)
- removed tiered pricing for Opus and Sonnet 4.6
- simplified side navigation
- simplified project list selector; shows all data sources
Fixed
- cmd window flash in Windows
- dark mode for API status page
- automatic internal converstation summaries using invalid UUID
[0.11.11] - 2026-03-05
Added
- custom toolset created when adding MCP server
- copy button to code blocks in message history
- auto-create custom tool set when adding MCP server
Changed
- content and styling of ‘BB app required’ notifications
- re-enabled support for cache status indicator - disabled by default
Fixed
[0.11.10] - 2026-03-02
Added
Changed
- moved collaborations list to side navigation
- moved project selector to side navigation
- redesigned Settings and Projects pages
- performance and reliability improvements
- improved layout and design
- analytics support
Fixed
- directory traversal and file listing in google datasource
- reactivity of metadata in collaboration and project headers
[0.11.9] - 2026-02-26
Added
- support for org-supplied Google app credentials
Changed
- minor fixes to data source dialog
- minor updates to onboarding tour flows
Fixed
- handling of item and folder paths for google data source
- refresh token handling for org-supplied Google app credentials
- handling of Gemini thought tokens
- updated features/capabilities for Gemini models
[0.11.8] - 2026-02-21
Added
- bulk onboarding for shared team workspaces
- support for remote chromium when using fetch_webpage or fetch_screenshot tools
- help dialog for RUT telemetry
- onboarding and help tours
- Added Claude Sonnet 4.6
- Added Google Gemini 3.1
Changed
- chat input design and responsiveness
- layout of home page and side navigation
- design of chat status
- upgraded
markedto v17 - fetch screenshot tool can grab sequence of images from whole page
- fetch screenshot tool can grab single element of page
- fetch web page tool will return available selectors for fetch screenshot
Fixed
- personal workspace in shared teams
- display of hierarchal workspaces (multiple orgs/teams)
- missing user preferences
- MCP servers using HTTP (remote) transport expecting OAuth config
- tool use error when no datasources are configured
[0.11.7] - 2026-02-12
Added
- Toolbar button to add data source to current project
Changed
- automated onboarding for organizations
Fixed
- hub selection via URL params
- stale object data
[0.11.6] - 2026-02-11
Added
Changed
Fixed
- Fix race condition for oauth flows
[0.11.5] - 2026-02-10
Added
Changed
- layout of Chat prompt input
Fixed
- features check for storage limits
- Oauth redirect URL
[0.11.4] - 2026-02-09
Added
- re-enabled support for reading docx files in filesystem datasource
Changed
Fixed
- subscription features check for organizations
- missing previous turn input_has for RUT telemetry
[0.11.3] - 2026-02-08
Added
- Support for Claude Opus 4.6 and ChatGPT 5.2
- Panel to view memories
Changed
- MCP servers scoped to team workspace
Fixed
- Loading of tools from MCP servers
- setting path for file uploads
[0.11.2] - 2026-01-31
Added
- option for custom origins when using local (LAN) hubs
Changed
- Improved handling of hub selection process
Fixed
- loading of organizations and workspaces
- display of workspace selector
- credits auto-topup not triggering
[0.11.1] - 2026-01-29
Added
- more support for teams and organzations with shared workspaces
- UUID display and copy component
Changed
- use more intuitive and less intrusive side nav expand/collapse button
- BUI layout tuning
- credit card not required for 100% discount coupons
- filtering for tools in local vs cloud hubs
Fixed
- auth token validation for forgot-password page
- handling of websocket protocol header for auth
- replace KV database with new name
- subscription upgrades with coupons getting free credits
- reset workspace selection when logging out
- image pasting (uploads)
- various billing and subscription fixes for teams & orgs
[0.11.0] - 2026-01-20
BREAKING CHANGES
π¨ IMPORTANT: Breaking Changes to project and collaboration storage. This Beyond Better upgrade will migrate all your projects and collaborations to cloud storage. Large projects, or many projects, can take 10 minutes or more to migrate. Allow sufficient time for migration to complete before proceeding with the upgrade. The move to cloud storage allows you to access your projects from any device. The changes also lay the foundation for teams and multiple workspaces.
IMPORTANT: Authentication changes
Due to underlying changes in authentication, you will need to log in to BB again after this update.
π Introducing RUT Telemetry
Beyond Better now tracks metrics to measure LLM “coherence”. Research shows LLM’s fail to follow instructions 52% of the time. RUT helps us detect and prevent conversation drift before it becomes obvious, leading to better AI responses for everyone.
This is foundational work to allow BB to be proactive when LLM begins to drift away from the conversation focus. The RUT technology is an advanced system for measuring how well the LLM is maintaining focus with the stated objective for the conversation. The system is only tracking metrics and this point. We need to collect data before we can enable proactive steps to keep LLM on track. If you prefer not to participate in the RUT telemetry; disable the “Conversation Coherence Monitoring” in Privacy Settings.
π Introducing Import and Export [BETA]
Beyond Better now supports exporting and importing workspaces or individual projects. The feature is currently in beta so please report any issues to support.
Added
- Multi-user and support for teams with multiple workspaces
- Support for choosing local or cloud Hub, with advanced config for custom Hub
- RUT telemetry - initial support for tracking LLM “coherence” to ensure focus and reduce conversation drift
- Ability to re-order projects with drag and drop.
- Export/import for workspaces and projects
Changed
- Projects, collaborations and conversation history stored in the cloud
- Support for local storage (contact support for details on migrating to local storage)
- Renamed API to Hub
- Renamed BUI to Web App
- Redesign of status updates in chat window
Fixed
- Authentication between web app and hub
[0.10.3] - 2025-12-25
HOTFIX for config migration
Added
Changed
Fixed
- don’t set undefined clearAtLeast
[0.10.2] - 2025-12-24
HOTFIX for Deno regression
Added
Changed
Fixed
- reverted to deno v2.5 to fix regression with macOS binaries
[0.10.1] - 2025-12-14
Added
- system prompt supports custom tool instructions
Changed
- use custom BB memory tool for all providers; supports memory searching and scope control
- use custom checkbox toggle
- additional model details for dynamically loaded models (specifically Ollama)
Fixed
- fixed variable evaluation in user-supplied system prompts
- loading of models and capabilities in BUI
- saving project configs for context management and memory
- resolving LLM provider for all models, including Ollama hosted models
- using Ollama models no longer requires local-only mode
[0.10.0] - 2025-12-08
π Introducing Intelligent Memory & Context Management
Beyond Better now remembers what matters. Our new memory system gives your AI assistant a persistent, searchable memory that works across all models and grows smarter with every interaction.
Multi-Level Memory Architecture
- Global memories - Knowledge that persists across all your projects
- Project memories - Context specific to each codebase
- Collaboration memories - Insights from individual conversations
The AI automatically manages its own memory, deciding what to save and recall. You can also take controlβadd, modify, remove, or search for specific memories anytime.
Automatic Context Management (Claude models)
Never lose your conversation flow. When conversations exceed 160k tokens, Beyond Better intelligently preserves important memories while archiving older content, keeping your sessions running smoothly without hitting token limits.
Added
- Support for memories managed by LLM
- Support for context management (compaction)
- Updated models for Grok and DeepSeek
Changed
Fixed
- dynamically update tools list when adding/removing MCP server and BB plugins
- updating tool results in message log entries
- pasting files into chat
[0.9.18] - 2025-11-29
Added
Changed
- combined tool input & output to single entry
- configurable tool bar
Fixed
- model selector list shows correct models
[0.9.17] - 2025-11-25
Added
- new models from Anthropic, OpenAI and Google
Changed
- include version number with display of tool names
Fixed
- show tool display name rather than tool ID
[0.9.16] - 2025-11-22
Added
- selected toolsets included in tool info panel
- live updates for config changes
- added config metadata (descriptions, restart requirements, etc)
Changed
- better control of debug logging at runtime
Fixed
- plugin loading for compiled javascript
- build process for plugins
- references to newer anthropic models
[0.9.15] - 2025-11-15
Added
- builtin and custom tool sets (collections of tools)
- UI to create custom tool sets
- UI to select toolsets for global or project settings
- short ‘purpose’ for tools (for display in UI)
- tool info panel in conversation header
- additional examples for description of most tools
- adding builtin tools to relevant tool sets
- captcha on signup and contact pages
Changed
- single prompt cache point for anthropic conversations
- per-tool config field in UI
- support both builtin and plugin tools with same name
- config version bumped to 2.2.2
Fixed
- auto-install bb-plugins in Windows and Linux
- delete empty sheet when creating new Google Sheet
- loading tools on every instance of tool manager
- live parsing of YAML for tool configs
[0.9.14] - 2025-11-04
Added
- html content for copy of message entries
- auto-install plugins
Changed
- tool descriptions with examples from metadata
- tidier layout of collaboration header
Fixed
- type interfaces for bb-tools support
[0.9.13] - 2025-11-01
Added
- added Claude Haiku 4.5
- user feedback and approval system
- notifications and progress (from mcp and internal tools)
Changed
- reformatted thinking blocks
- api <-> bui communication uses JSON-RPC format
- extended mcp tool call timeout
- extended LLM request timeout
- init collaboration when switching collaborations
- additional notification types
- content for collapsed user menu
- extra help text for new collaboration page
Fixed
- test case reliability and coverage
- loading of ‘last used’ project editor
[0.9.12] - 2025-10-14
Added
Changed
Fixed
- replaceAll for searchReplace in filesystem datasources
- missing tool_results when MCP server fails to return tool listing
- reconnection handling for http MCP servers
- auth login session not being saved
- landing page requiring auth session
- permissions for oauth flow in DUI
- autoscroll behaviour for chat messages list
[0.9.11] - 2025-10-08
Added
- readonly option for datasources
Changed
Fixed
[0.9.10] - 2025-10-04
Added
- support resource autocomplete suggestions for all datasource types
- option to exclude datasource from autocomplete suggestions
- UI controls to filter list of message entries
- new tool for download_resource
- support for remote MCP servers
- support for MCP oAuth
- support for MCP sampling
- support for ‘best model’ selection for sampling requests
- foundation for session management - multi-user in API, with auth from bui/cli
Changed
- updated default model to Claude Sonnet 4.5
- fetch timeout for bbLLM
- updated deps for MCP SDK
- UI for MCP server configs
Fixed
- reading PDF documents with Claude
- project/request context in McpManager
- reconnection for MCP sessions
[0.9.9] - 2025-09-05
Added
- [saas] Google File Picker for restricted access to drive contents
- Conversation export and copy to clipboard
Changed
- filtered instructions for datasources
- additional datasource instructions for utility tools (rename, remove, etc)
Fixed
- [saas] file rename for google datasource
[0.9.8] - 2025-08-31
Added
- [saas] google auth flow in app (dui)
Changed
Fixed
- searching collaborations
[0.9.7] - 2025-08-30
Added
- [saas] support for Google Sheets editing
Changed
- [saas] refactored google datasource for better extensibility
Fixed
[0.9.6] - 2025-08-17
Added
- deep research (using new web_search tool)
Changed
- [saas] google datasource supports resource rename and delete
- [saas] google datasource supports loading non-doc resources (eg image files)
- [saas] google and notion datasources support more robust resource searching
Fixed
- save message history after fatal conversation errors
- caching for feature access checks
[0.9.5] - 2025-08-14
Added
- support for pricing tiers for models with large context windows
- UI feedback for tiers in progress bar
Changed
- Claude Sonnet uses 1 million token context window
Fixed
- updated details for capabilities & pricing for Anthropic, Google and OpenAI models
[0.9.4] - 2025-08-12
Added
- support for Claude Opus 4.1
- support for ChatGPT 5 (& mini & nano)
Changed
Fixed
[0.9.3] - 2025-08-12
Added
- datasource instructions for writeResource content formats
Changed
- less verbose system prompts (rely on datasource instructions instead)
Fixed
- [saas] google datasource supports structured content for writeResource tool
- ensure terminal detachment during app self-update
- [saas] invalidate stale auth config after updating google auth in BUI
[0.9.2] - 2025-08-11
Added
Changed
Fixed
- [saas] oAuth flow targets and config
[0.9.1] - 2025-08-10
Added
Changed
- additional editing instructions for different datasources
Fixed
- [saas] apply range edits in descending order to avoid index shifting with multiple operations
[0.9.0] - 2025-08-09
Notes
Major Upgrade
Beyond Better now supports both read and write editing for external data sources. Notion and Google Docs
are currently supported with more providers coming soon [saas]. See the “Data Sources” help page to learn how to
add data sources and configure authentication.
Added
- [saas] read/write support for editing Notion documents
- [saas] read/write support for editing Google documents
- dynamic loading for third-party datasources
- integration tests for datasources
Changed
- remaining tools updated to use datasource accessors
- deprecated searchAndReplace and rewriteResource tools in favour of editResource and writeResource
- support for contentFormat and editTypes to all resource tools
- releases moved to BB storage bucket
Fixed
- save in-progess prompt when navigating within app
- validate content parts returned from LLM
[0.8.15] - 2025-07-27
Added
Changed
Fixed
- undefined collaboration in header for new conversations
[0.8.14] - 2025-07-27
Added
- coupon discounts for subscriptions
Changed
Fixed
- forgot password reset via email link
- prorated amounts for plan upgrades
- crash following app self-update
[0.8.13] - 2025-07-21
Added
- config option to allow project root external symlinks
Changed
- follow symlinks when finding git root
- follow symlinks when searching file suggestions
- updated subscription plan cards with promo/discount fields
Fixed
- model access in local-only mode
[0.8.12] - 2025-07-16
Notes
This is a quick fix. See release notes for v0.8.11 for full list of recent changes.
Beyond Better has a major update to subscription plans and billing system. See Pricing page for more details.
BREAKING CHANGES
- The billing system has been completely replaced to support new subscription plans. There should be no risk of breaking local project configs, but users should be aware of new subscription plans and token credit system.
Added
Changed
- added flag for checking model access
Fixed
[0.8.11] - 2025-07-16
Notes
Beyond Better has a major update to subscription plans and billing system. See Pricing page for more details.
BREAKING CHANGES
- The billing system has been completely replaced to support new subscription plans. There should be no risk of breaking local project configs, but users should be aware of new subscription plans and token credit system.
Added
- New subscription plans and billing system
- Features system to support new plans
- Credit auto top-up option
- Invoice history and token usage tab in Settings
Changed
- Plans and Tokens tab updated for new billing system
Fixed
- Billing calculation errors
[0.8.10] - 2025-07-04
Added
Changed
- switched domain to beyondbetter.app
Fixed
- catching global errors in the api
[0.8.9] - 2025-06-28
Added
- Edit conversation title
- Starred (favourite) conversations
Changed
Fixed
[0.8.8] - 2025-06-28
Added
- self-update for app
Changed
- Changed from MIT license to GNU Affero General Public License v3.0
Fixed
[0.8.7] - 2025-06-27
Notes
This is a quick fix. See release notes for v0.8.6 for full list of recent changes.
Added
Changed
- formatting and parsing of release notes
Fixed
- loading and using assistant and person name from project config
- token usage progress display was doubled
[0.8.6] - 2025-06-27
Notes
This is a quick fix. Including change notes from v0.8.5 as well.
BREAKING CHANGES
- Major refactor to storage format for conversations
Added
- refactored storage format as foundation for improved multi-agent support, model selection, and token reporting - new Collaboration class
- model selection for all AI roles (orchestrator/agent/admin)
- backup project before migration
- Enhanced release notes system with critical notices
- Release notes parsing from GitHub releases
- Breaking changes detection in DUI
Changed
- progress bar for token usage of context window
- validate model config options
- chat input receives focus when creating new conversation
- GitHub release workflow now extracts changelog information
- DUI update prompts show enhanced release information
Fixed
- updates for token and stats reporting in BUI
- passing model config between API and BUI
- o3 model params
- cleanup duplicate tests
- cleanup storage migration functions
- validation for extended thinking budget in global project defaults
- ensure allowedCommand for run_command tool has at least one command
- ensure user and assistant name are loaded from project config
[0.8.5] - 2025-06-27
BREAKING CHANGES
- Major refactor to storage format for conversations
Added
- refactored storage format as foundation for improved multi-agent support, model selection, and token reporting - new Collaboration class
- model selection for all AI roles (orchestrator/agent/admin)
- backup project before migration
- Enhanced release notes system with critical notices
- Release notes parsing from GitHub releases
- Breaking changes detection in DUI
Changed
- progress bar for token usage of context window
- validate model config options
- chat input receives focus when creating new conversation
- GitHub release workflow now extracts changelog information
- DUI update prompts show enhanced release information
Fixed
- updates for token and stats reporting in BUI
- passing model config between API and BUI
- o3 model params
- cleanup duplicate tests
- cleanup storage migration functions
[0.8.4] - 2025-06-13
Added
- use custom llm-proxy service without strict resource constraints
Changed
- Redesign model selection and help
- Use SVG instead of emoticons for provider and model capabilities icons
- Add version number to /api/v1/status
- rewrote process restart strategy
Fixed
- file logging for BUI
- request headers in proxy in DUI
[0.8.3] - 2025-06-11
Added
Changed
Fixed
- notifications from app
- subscription billing
[0.8.2] - 2025-06-10
Added
- model selection for projects and conversations
- model combinations easy select
- list of model system cards
- support for Gemini, OpenAI, Groq
- notifications when conversation statement completes
- support for dark/light/system theme
- auto-save prompt content between page reloads
- show thumbnails in chat history for uploaded files
- sync model data to llm-proxy
Changed
- flexible token-type for model pricing (various caching types, extensible for new token types)
Fixed
- duplicate conversation ID when creating new conversation in BUI
[0.8.1] - 2025-05-29
Added
- return contextual content for find resource tool
Changed
- moved handling of llmProvider from controller to interaction
- UI updates and improvements
- resource auto-complete (suggestions) include data source name
Fixed
- scrolling behaviour for chat history when prompt input is modified
- model selection and defaults
[0.8.0] - 2025-05-24
BREAKING CHANGES: This release removes ‘project root’ in favour of ‘data sources’; moves
location of project config; changes format of storage and configurations; and more.
IMPORTANT: Backup your project .bb and .config/bb directories before upgrading.
Added
- Support for Claude v4 (Sonnet and Opus)
- Support for Data Sources - major refactor across whole code base
- [BETA] Support for Notion data source
- Reading MCP resources
- Use Google Gemini models on local-only mode
Changed
- System prompt lists data sources rather than showing file listing
- Tools use data sources instead of project root
- Updated content for home page
- Layout and design improvements
- Moved all project config and data to shared bb directory
Fixed
- various fixes and improvements during refactor
[0.7.5] - 2025-04-10
Added
- inspector for API internals
- check for existing user during signup
- option to re-send email verification message
Changed
- improved error messages on login auth failures
- API log formatting
Fixed
- don’t fail on already verified email addresses
[0.7.4] - 2025-04-07
Added
- extra fields included in signup form
Changed
- enabled streaming for anthropic requests to allow larger max output tokens
- deprecated ‘git’ project type
- switched to internal git library instead of CLI wrapper
Fixed
- saving payment method when changing plans
[0.7.3] - 2025-04-02
Added
- log paths for DUI and proxy
- button in DUI to open log files
Changed
- instructions and styling for server control toggle in DUI
- retry/fallback for http/s protocol switching for API
- moved all log paths in DUI to separate section
Fixed
- external links in DUI open in default browser
- control of internal proxy
- ensured status in ChatInput is fully reactive
[0.7.2] - 2025-03-31
Added
Changed
Fixed
- fixed handling of logDataEntry in CLI
[0.7.1] - 2025-03-30
Added
- Orchestrator/Agent task delegation
- New tool for image manipulation
- Log version number during API/BUI startup
Changed
- design overhaul for messages in chat window
- system prompt for orchestrator and agent conversations
- display and formatting for nested agent tasks in BUI
Fixed
- removed use of primaryInteractionId
- naming of logEntry vs logDataEntry
- cache tools list from MCP servers
[0.7.0] - 2025-03-23
Added
- Added support for MCP servers, with UI configuration panels
Changed
Fixed
- whitespace before pre/code blocks
[0.6.18] - 2025-03-20
Added
- Starter prompts for new conversations
- Support for pasting images into conversations
Changed
- Styling for login and signup pages
Fixed
- Fixed value too large error when caching large LLM responses
[0.6.17] - 2025-03-17
Added
- Support for Claude 3.7 and extended thinking (reasoning)
- Model capabilities system
- Passing LLM response metadata to frontend (BUI)
- Config panel for controlling conversation options
- Info panel for current model details and settings
Changed
- Improved display of thinking process extracted from LLM
- Structured metadata provided to LLM for each statement and tool turn
Fixed
- Fixed display of tokens counts for statements vs conversation
[0.6.16] - 2025-02-23
Added
- Provider support for Google Gemini [beta]
Changed
Fixed
- switched to custom confirmation dialogs to fix missing webview native dialogs
- rewrote window size & positioning logic and state handling - respects multiple displays
[0.6.15] - 2025-02-18
Changed
- tool descriptions
Fixed
- change glob to regex lib to fix windows path globbing
[0.6.14] - 2025-02-17
Added
- Signed and notarized app for macOS
Fixed
- Windows installer and logging config
[0.6.13] - 2025-02-16
Fixed
- stopped signup verification looping
[0.6.12] - 2025-02-16
Fixed
- landing page auth check
- project config migration backups
[0.6.11] - 2025-02-07
Fixed
- validation for global config and new config missing defaultModels
- tokenUsage for chats saved with conversation
- token savings calculations
[0.6.10] - 2025-02-01
Added
- created base class abstraction for OpenAI compatible providers
- added support for Ollama
- added support for DeepSeek
- conversation and project migrations for version upgrades
Changed
- llmKeys moved under llmProviders
- run_command tool supports head/tail truncation for stdout/stderr
Fixed
- token calculations and reporting
[0.6.9] - 2025-01-23
Changed
- external link display in dui
Fixed
- default port for bui changed in configs
- dui handles ‘corrupt’ version strings from api
- file loading in api
- saving project type updates project config
[0.6.8] - 2025-01-21
Added
- new tool for display file
Fixed
- fix for saving project config after edit in project settings
- tool config shows blank text area instead of blank json
[0.6.7] - 2025-01-18
Changed
- Project editor has more configuration fields
Fixed
- Propagating errors from LLM
[0.6.6] - 2025-01-16
Fixed
- force a release for BUI build
[0.6.5] - 2025-01-16
Added
Changed
- layout for project manager
Fixed
- version reporting for api
[0.6.4] - 2025-01-13
Fixed
- Chat Input regressions
- Navigation in Project Manager
- Propagating errors from LLM
[0.6.3] - 2025-01-12
Fixed
- valid semantic version to keep Tauri happy
[0.6.2a] - 2025-01-12
Fixed
- default config for defaultModels
[0.6.2] - 2025-01-12
Added
- Refresh button for Subscription usage
- Prompt history in Chat Input
Changed
Fixed
- Location for log and run files in Linux
- Subscription page updates when navigating away and back
[0.6.1] - 2025-01-12
Added
- Settings in BUI for project defaults
- [beta] Support for OpenAI provider (needs image_url instead of b64 data)
- [alpha] Support for DeepSeek provider (not returning tool_calls message) (via llm-proxy only)
Changed
- DUI control for API & BUI
- LocalMode settings in DUI for API & BUI
- Redirect from login page if already logged in
Fixed
- Changed KV key to fix auth session restoration in API
[0.6.0] - 2025-01-08
Added
- user account with subscription plans
- llm proxy to remove requirement for api key
- support for external tools
- dedicated log (per conversation) of LLM request/responses
- added remove_files tool
- binary build for BUI
Changed
- docs shuffle
- updated deps (including Anthropic SDK with GA prompt caching)
- improved prompting to encourage use of request_files tool
Fixed
prebackground in dark mode- auth handling in BUI
[0.5.12] - 2024-12-20
Added
- http/websocket proxy in DUI to work around mixed content warnings in webview
- persisting windows size/position
- dark mode support
[0.5.11] - 2024-12-18
Added
- Conversation selector on chat page
Changed
- Project Manager links direct to chat page
- Collapsed conversation list on chat page
- Improved conversation metadata
[0.5.10] - 2024-12-16
Fixed
- Fixed imports for Windows build of desktop app
[0.5.9] - 2024-12-16
Changed
- Modified API process handing in Windows version of desktop app
[0.5.8] - 2024-12-16
Fixed
- Automatic installation of binaries using Windows desktop app
[0.5.7] - 2024-12-16
Added
- config option to set project dir for API control
Changed
- updating logging in DUI
[0.5.6] - 2024-12-15
Added
- add CSP header to BUI
[0.5.5] - 2024-12-15
Fixed
- Install prompt in desktop app
[0.5.4] - 2024-12-15
Fixed
- workflow syntax
[0.5.3] - 2024-12-15
Fixed
- GitHub workflow version handling
[0.5.2] - 2024-12-15
Added
- Chat page added to desktop app (when using api in secure/tls mode)
- Debug mode in desktop app to allow using localhost browser interface
Changed
- Desktop app changed to universal binary for macOS
Fixed
- projectId (& config) is optional for many CLI commands
- API starts automatically when launching desktop app
- Config in desktop app verified before starting API
- API url generation in browser app
- Default values set for global config
[0.5.1] - 2024-12-11
Changed
- Conversation counts moved to metadata bar
Fixed
- BB Desktop logs to standard log diretory
- BB Desktop looks for binaries in expected location
- Browser handles project selection
- Changing conversations is fixed
[0.5.0] - 2024-12-09
Added
- Multi-page design for browser interface
- Automatic download of BB app
- App to manage the BB Server
- Auto-install/upgrade of bb and bb-api
- Shows installed BB API version
- Automatic conversation summary
Changed
- Global and Project configs schema (v2)
- Use projectId instead of startDir to identify projects
- Moved prompt cache points to all user messages, not just request_file tool use
Fixed
- Typeguards for tool results
- Lots of display fixes for missing metadata and token usage
[0.4.1] - 2024-12-02
Changed
- messages sent with cmd/meta-return; new lines with return
Fixes
- scroll indicator shows correct unread count
- auto-scrolling disabled when page is not at end
- file suggestion triggers are less keen
Added
- support for bb doctor
[0.4.0] - 2024-12-02
Changed
- added file listing autocomplete
- added version display in BUI, and check and warning for minimum API version
- added API auto-upgrade
- added API endpoint to initiate upgrade from BUI
- added ‘migrate’ CLI command to change from system to user-installed binaries
- added ‘upgrade’ CLI command to upgrade to latest version of
bbandbb-api - standardised log entry formatting with subtitle and preview components
- combined prompt/answer auxiliary log entries
- added common formatting functions and styles for tools
- improved log entry metadata display
- disabled auto-scroll if not at bottom of the page - added scroll to bottom button
[0.3.11] - 2024-11-26
Changed
- normalized token usage, conversation stats vs metrics;
- fixed token calcs for statement and conversation
- extract tool thinking with every turn
- all usage tokens sent to bui
- fixed indents for new global config file
[0.3.10] - 2024-11-26
Changed
- fixed indentation for default global config file
[0.3.9] - 2024-11-25
Changed
- fixed name of ca cert when adding to root trust store
- extended wait time for API start
- added IDLE status to progress handler switch
[0.3.8] - 2024-11-25
Changed
- Fixed status handling to force immediate IDLE status after conversation answer, preventing stuck progress indicators
[0.3.7b] - 2024-11-24
Changed
- must await
[0.3.7a] - 2024-11-24
Changed
- Updated Github actions to use Deno v2
[0.3.7] - 2024-11-24
Changed
- Upgraded to Deno v2
- Fixed status queue processing to ensure IDLE status is always displayed after conversation answer
- Fixed error loading global config
- Documentation updates
Added
Improved TLS certificate management:
- Automatic certificate generation and trust store integration
- New
bb securecommand for certificate management - Platform-specific trust store handling
- Removed external mkcert dependency
Enhanced security documentation:
- New certificate management guide
- Trust store documentation
- Security troubleshooting guide
- Platform-specific certificate handling
- Browser compatibility guidance
Improved status page:
- Enhanced certificate information display
- Trust store status
- Platform-specific guidance
- Browser warning solutions
[0.3.6] - 2024-11-23
Changed
- Improved status handling in BUI and CLI to provide smoother transitions and more reliable idle state handling
- Updated conversation_metrics and conversation_summary tools to provide clearer guidance about stopping after completion unless explicitly asked to continue
[0.3.5] - 2024-11-23
Changed
- Fixed API startup failure recovery mechanism
- Added graceful handling of tool cancellation states
- Implemented error message propagation from LLM to BUI/CLI
- Added distinct status indicators for Claude, API, or tool execution
- Exposed prompt cache status via API websocket
- Added support for multiple git repositories in subdirectories
[0.3.4] - 2024-11-20
Changed
- fixed bug with missing data dir
- fixed bug with symlinks in project listing
[0.3.3] - 2024-11-19
Changed
- changed search_project tool to use Deno-native
walkoptions for match and skip - updated tool description to clarify use of ** vs * in file patterns
[0.3.2] - 2024-11-18
Changed
- improved websocket reconnection handling
- improved conversation admin (adding and deleting)
- improved notifications
- prompt caching indicator
[0.3.1] - 2024-11-17
Changed
- added help dialog
[0.3.0a] - 2024-11-17
Changed
- hotfix for broken input
[0.3.0] - 2024-11-17
Changed
- made some changes to BUI
[0.2.6] - 2024-11-10
Changed
- refactored handling (recording and reporting) of token usage
- added tests for persistence layer
- changed from adding files inside
tags to using separate content part - changed handling of content extraction from LLM response
- restructured contents of docs directory
- improved parsing for file mime type
[0.2.5] - 2024-11-08
Changed
- added ‘conversation_summary’ tool
[0.2.4] - 2024-11-06
Changed
- set apiUseTls to false during project init if certs could not be created
- fixed message hydration to keep last two versions of each file
- changed prompt caching to add cache points of three most recent messages with hydrated files
- removed “statement objective” on first statement, in favour of “conversation objective”
- fixed statement counts in messages.jsonl
[0.2.3] - 2024-11-03
Changed
- changed one-shot task to use text format as default; json format is cli arg
- updated cli config sub-command to work with full, global, and project level configs
- updated cli config sub-command to accept more updates
- updated cli config sub-command to output in colour
[0.2.2] - 2024-10-30
Changed
- task objectives to help Claude focus
- tool tracking (progress)
- fixed unwanted cache busting with updated project details
- stopped creation of conversation storage for ephemeral chats
[0.2.1] - 2024-10-29
Changed
- added config option and CLI arg for maxTurns
[0.2.0] - 2024-10-23
Changed
- Full rewrite of tool descriptions (based on guidance from Anthropic’s “computer use” tools)
- Updated to latest Sonnet (WOW, I’m impressed)
[0.1.6] - 2024-10-22
Changed
- changed search_project tool to accept compound file patterns eg. .js|.ts|*.json
[0.1.5e] - 2024-10-17
Changed
- version bump to test github workflow releases
[0.1.5d] - 2024-10-17
Changed
- version bump to test github workflow releases
[0.1.5c] - 2024-10-17
Changed
- version bump to test github workflow releases
[0.1.5b] - 2024-10-17
Changed
- version bump to test github workflow releases
[0.1.5a] - 2024-10-17
Changed
- version bump to test github workflow releases
[0.1.5] - 2024-10-17
Changed
- more fixes for Windows release
[0.1.4a] - 2024-10-17
Changed
[0.1.4] - 2024-10-17
Changed
[0.1.3a] - 2024-10-17
Changed
- version bump to test github workflow releases
[0.1.3] - 2024-10-17
Changed
- fix Github workflow actions
[0.1.2] - 2024-10-17
Changed
- Added BB Manager convenience scripts
- changed location of .bat files
- Updated system prompt, conscious thinking
[0.1.1] - 2024-10-16
Changed
- always include LLM response in answer, regardless of ‘thinking’ text block
[0.1.0] - 2024-10-16
Changed
- renamed BBai to Beyond Better
- renamed cli tools
- renamed all code references and deploy workflows
[0.0.26-beta] - 2024-10-14
Changed
- normalized tool results and responses - better feedback to LLM and clearer output in conversation
- New LLM tool:
conversation_metricsfor calculating the current turn and token counts,
needed for the upcomingconversation_summary(& truncation) tool - improved handling for conversation log entries; loading saved conversations and conversation answers
have the same handling as normal entries during conversation
[0.0.25-beta] - 2024-10-10
Changed
- added new tool MultiModelQuery
- updated tests to load tools via tool manager
- moved tool configs to user config
- args for tool constructors set from metatadata json
[0.0.24a-beta] - 2024-10-07
Changed
- hotfix for compile target arg
[0.0.24-beta] - 2024-10-07
Changed
- generate tools manifest file to facilitate loading from compiled binary
- dynamically include tools at build compile time
- update build scripts and release workflow to use compile script
[0.0.23-beta] - 2024-10-06
Changed
- Overhaul of tool packaging and loading - each tool is self-contained for easier sharing
- Improved conversation entry formatting, including inline images
- Safely ignore cert warnings for localhost
- New LLM tool:
move_filesfor moving files and directories to a new directory within the project - New LLM tool:
rename_filesfor renaming files and directories within the project
[0.0.22-beta] - 2024-10-04
Changed
- Upgrade deps to use latest Astral (fix for fetch screenshot tool)
[0.0.21a-beta] - 2024-10-01
Changed
- Hotfix for incorrect method name to create TLS cert
[0.0.21-beta] - 2024-10-01
Changed
- Copy rootCA.pem for deno’s fetch custom http client
- Fix for check of existing certs
- Silenced file missing exception
[0.0.20-beta] - 2024-10-01
Changed
- Fix for nested array in tool results
- Improved mime-type checking
[0.0.19-beta] - 2024-09-30
Changed
- Added option for regex patterns to search and replace tool.
[0.0.18-beta] - 2024-09-29
Changed
- Changed BUI to prefer StartDir from URL
- Changed search project tool to default to case-insensitive and ignore git when searching
[0.0.17-beta] - 2024-09-28
Changed
- Refactored search_project tool to use native stream reader with buffer and native regex, rather than external grep
[0.0.16-beta] - 2024-09-27
Changed
- Fixed ‘git not found’ error when running init
- Updated
bbstrings tobb.exefor Windows - Fixes for Windows compat
- Added config option for API and BUI to listen with TLS
- Updated CLI and BUI to initiate TLS connections if useTls is set
- Added certificate generation to init process
- Updated docs
[0.0.15-beta] - 2024-09-25
Changed
- Refactored Split config handling; global, project, full
- Added config for apiHostname
- Changed BUI to also load apiHostname and startDir from URL
- Fixes for init wizard
- Added support for image files to request_files tool
- Added support for prompt caching to full message history
- Easy install; Windows MSI and one-liner for macOS and Linux
[0.0.14a-beta] - 2024-09-14
Changed
- Release builds for each platform
- Updated docs for INSTALL and README
[0.0.13-beta] - 2024-09-14
Changed
- Refactored Split config handling; global, project, full
- Added config for apiHostname
- Changed BUI to also load apiHostname and startDir from URL
- Fixes for init wizard
- Changed tests to create configured project for each unit
[0.0.12b-beta] - 2024-09-08
Changed
- Hotfix-2 for over-ambitious api client
[0.0.12a-beta] - 2024-09-08
Changed
- Hotfix for over-ambitious api client
[0.0.12-beta] - 2024-09-08
Changed
- Added wizard for
bb init; will re-use existing config values for defaults if present - Changed git to be optional
- Split config handling into global and project
- Improved handling for API control (port number)
[0.0.11-alpha] - 2024-09-06
Changed
- BUI is not a hosted site (can still be run manually from localhost)
- Running
bb startwill open browser page for hosted BUI, with apiPort pointing to localhost API
[0.0.10b-alpha] - 2024-09-03
Changed
- Hotfix for bui
[0.0.10a-alpha] - 2024-09-03
Changed
- Hot fix for deno.lock and build command
[0.0.10-alpha] - 2024-09-02
Changed
- Added custom tool content formatters for console and browser
- Added browser interface
- Added prompt caching for tools and system prompt
- Updated cli tool to auto-start api server
[0.0.9-alpha] - 2024-08-18
Changed
- Added mock stubs to bypass calls to LLM during unit tests
- Added web page fetch tool
- Added web screenshot fetch tool
- Reworked tool results, tool response, bb response for better conversation logging
- Refactored finalize callback for tool runs
- Cleaned up dangling event listeners
[0.0.8a-alpha] - 2024-08-15
Changed
- Hot fix for missing conversation log entries
[0.0.8-alpha] - 2024-08-15
Changed
- Foundations for Orchestrator/Agent structure
- More tests
- Console logging cleanup
- Error handling
[0.0.7-alpha] - 2024-08-05
Changed
- Added websocket support for live updates of conversation logging
- Added event manager
- Added manager for project editors
- Improved typescript type handling
- Refactored terminal console handling
[0.0.6-alpha] - 2024-08-03
Changed
- Move tools to a ToolManager class
- Migrate each tool to dedicated class
- Fixes for gathering ProjectDetails
- Improved conversation logging
- More reliable file hydration in messages
- Better tool input validation and handling
[0.0.5a-alpha] - 2024-08-01
Changed
- Hot fix for multiple tool result blocks
- Hot fix for undefined usage tokens
[0.0.5-alpha] - 2024-08-01
Changed
- Added terminal support for multi-turn conversations
- Applied formatting to chat logs for easier reading
[0.0.4-alpha] - 2024-07-28
Changed
- Add workflow to automatically create a Github release and deploy to package managers
- Implement git commit after patching
- Create a class for a “fast” conversation with haiku (for git commit messages, semantic conversation titles, etc.)
- Use haiku to create semantic names for conversations based on initial prompts
- Use haiku to write commit messages
[0.0.3-alpha] - 2024-07-27
Changed
- Add support for Homebrew on macOS
- Lots of refactoring improvements for
- tool use
- conversations
- stateless requests (data persists across API restarts)
- file searching, adding to conversation, and patching
- project editor to handle different data sources (only local filesystem so far)
[0.0.2-alpha] - 2024-07-23
Added
- Initial project setup
- Basic CLI and API functionality
- File handling capabilities
- Conversation management features
[0.0.1-alpha] - 2023-07-20
- Initial alpha release