CREW ONE
Book an intro

Changelog / The product in progress

What changed. Why it matters.

New features, improvements and fixes from the Crew One team.

v0.13

Company Wiki & Connector Health

  • Company Brain
  • Connectors
  • Chief of Staff

Your crews can now read the company wiki and cite what they read, and your connections tell you when they break instead of failing quietly inside a workflow.

Ask your company wiki

The Chief of Staff now reads the pages your company writes down — handbook, metrics, decision log — and answers from them directly. Every answer names the page it came from, so you can check the source rather than trusting a paraphrase. Ask "what did we decide about vendor spend?" and you get the decision and where it was recorded.

Connectors that raise their hand

A broken connection used to surface as a workflow that quietly returned nothing. Connector health now tracks each connection and alerts you when one needs attention, so a lapsed token is something you fix in a minute rather than something you discover a week later in a briefing that was missing half its input.

Tokens that fail closed

Each connection resolves through a per-user rail. If your token cannot be resolved the call fails rather than falling back to a shared credential, so one person’s access is never quietly borrowed to do another person’s work.

Your Notion OKRs become Goals

If you keep objectives in Notion, they now sync into durable Goals that crews can read and track against. You keep writing them where you already write them, and the rest of the system stops being blind to what you are actually aiming at.

Commitments that close themselves

When a commitment is met, Crew One detects the resolution and closes it rather than waiting to be told. Anything that no longer matters, you can dismiss outright, so the open list stays a real list of what is outstanding.

Everything in this release
  • The Chief of Staff reads your company wiki and cites the page each answer came from (BAS-1626).
  • Durable connector health, with alerts when a connection needs attention (BAS-1419).
  • A per-user connector rail with fail-closed token resolution (BAS-1432).
  • Notion OKRs sync into durable Goals (BAS-1412).
  • Commitments close the loop, with detected resolution and a founder who can dismiss (BAS-1346).
  • Canonical entity resolution, so one person or company resolves to one identity across every surface (BAS-1416).
Fixes
  • The connector-health ledger is registered as per-user erasable data (BAS-1424).

v0.12

Goals, Reviews & the Operating Budget

  • Goals
  • Reviews
  • Governance
  • Budget

The weekly company Review became a real surface built on live data, Goals became objects that persist, and every run now has a cost and a stop button.

The weekly company Review

Once a week, Crew One assembles what moved, what stalled and what needs your judgement, then publishes and delivers it. It is built from live runtime data rather than a summary of a summary, so the Review reflects what actually happened rather than what was reported.

Goals that persist, and show what is at risk

Goals are now durable objects with their own read path and surfaces, instead of context that evaporates between sessions. An At-risk view derives from each goal’s review date, so the ones drifting out of attention surface on their own rather than waiting for you to remember them.

An operating budget with real limits

Every run writes to a durable cost ledger, and you can set spend limits per run and per company. You can see what the work costs before the invoice tells you, and a runaway job hits a ceiling instead of a bill.

Stop a run, and its approvals stop with it

You can pause or cancel a running execution. Anything it was holding for your approval is revoked at the same time, so cancelling means cancelled — not a stopped run with an approved action still queued to fire behind it.

Evidence that names its source

Evidence cites the connector it came from. When a source cannot answer, it says so plainly rather than returning a confident gap, so you can tell the difference between "nothing happened" and "we could not look".

Everything in this release
  • The weekly company Review is scheduled, published and delivered (BAS-1221).
  • Durable Goals — read route, proxy and wired surfaces — with an At-risk tab (BAS-1280, BAS-1284).
  • Company v2: five modules over live runtime data (BAS-1279).
  • A durable cost ledger, with per-run and per-company spend limits (BAS-1231).
  • Pause or cancel a running execution, revoking its pending approvals (BAS-1227).
  • A governed disposition command surface (BAS-1223).
  • Connector-citable Evidence, with honest source failure (BAS-1228).

v0.11

Proactive Crews & Connected Accounts

  • Crews
  • Integrations
  • Briefing
  • Slack

Crews began doing work on a cadence rather than only when asked, and started working from the tools your company already runs on.

Crews that run on a schedule

A proactive worker now runs in production, scheduling crew work and delivering the result to Slack. Your briefing arrives because it is due, not because you remembered to ask for it — which is the difference between a tool you operate and one that operates alongside you.

Connect Google, Slack, Notion and Linear

Each integration is authorised from your own account over OAuth, and configured per user rather than per workspace. Crews read the context they need from the systems you already work in, so you stop re-explaining your company to a blank chat box.

A briefing built from your real data

The morning briefing reads your actual calendar, your actual inbox and your real Linear tasks. It tells you what changed overnight, what needs your judgement and what can wait, drawn from the same sources you would have checked yourself.

Memory that consolidates

A consolidation pass runs monthly and quarterly, folding what happened into durable context rather than letting it pile up as raw history. Context compounds instead of accumulating, so the system gets more useful over time rather than slower.

Everything in this release
  • A proactive worker service — scheduler plus Slack delivery — running in production (BAS-544).
  • Scheduled crew briefs across eight teams (BAS-578).
  • OAuth connections for Google, Slack, Notion and Linear, authorised per user.
  • Per-user integration configuration, served from the product (BAS-524).
  • The briefing reads your real tasks from Linear (BAS-487).
  • Chat is served on the production app (BAS-649).
  • A memory-consolidation Curator, with monthly and quarterly digests.

v0.10

The Company Brain

  • Company Brain
  • Memory
  • Security

Company context became a queryable store that answers with citations — scoped to your company, not pooled across everyone.

Answers that cite their source

Ask the brain a question and the answer names where it came from, wired end to end through the provider adapters. A citation is what separates an answer you can act on from one you have to go and verify first.

Your context stays your context

Company isolation is enforced at the brain endpoint, so one company’s context is never served to another. This is the foundation the rest of the product is built on — memory is only useful if it is also contained.

Wiki and CRM, behind row-level security

Company knowledge and customer records live in tables with role grants and row-level security. Crews read what they are entitled to read, and the boundary is enforced by the database rather than by convention.

Everything in this release
  • Citations wired end to end through the provider adapters (BAS-146).
  • Company isolation active on the brain endpoint (BAS-144).
  • Wiki and CRM tables with role grants and row-level security (BAS-58, BAS-59, BAS-60).
  • Live decisions and crews read endpoints, with proxy routes (BAS-386).
  • Email verification wired through BetterAuth and Resend (BAS-340).

v0.9

Multi-Surface Gateway & Identity Linking

  • Gateway
  • Identity
  • Multi-Surface

Crew One now follows you across Slack, WhatsApp, and the web with unified identity and persistent memory.

  • Unified identity across Slack, WhatsApp, and web
  • Persistent memory that follows you across every surface
  • Session isolation for group chats vs direct messages
  • Smart activation modes - mention, always, or smart heuristics
Everything in this release
  • Cross-surface identity linking with automatic resolution
  • Gateway dispatches messages to the right agent team from any surface
  • Session types: main (1:1), group, and channel with isolated context
  • Lane-based queue with priority processing and backpressure
  • Callback-based metrics for decoupled observability
Fixes
  • Fixed session context bleed between group and direct messages
  • Resolved identity cache invalidation on link/unlink operations
  • Corrected message ordering in high-throughput gateway scenarios

v0.8

Plan-and-Learn Orchestration Engine

  • AI
  • Orchestration
  • Learning

Introducing PaL - the brain that routes your requests by complexity, learns from every interaction, and gets faster over time.

  • Complexity assessment on a 1-5 scale routes to the right strategy
  • Think-Execute-Analyse loop ensures quality before responding
  • Pattern learning with Bayesian confidence updates
  • Context budgeting prevents token waste on complex requests
Everything in this release
  • ComplexityAssessor classifies requests from Trivial (1) to Elaborate (5)
  • DAG-based execution plans for complex multi-agent requests
  • ACE-FCA: explicit Research/Plan/Execute phases with context compaction at 70%
  • Skill-aware routing matches requests to curated per-team domain skills, backed by a 218-skill library
  • Memory flush at 65% context usage for pre-compaction persistence
  • Commitment tracking extracts and monitors promises made in conversations
Fixes
  • Fixed plan executor hanging on circular dependencies in DAG
  • Resolved context budget miscalculation for nested agent calls
  • Corrected pattern store confidence decay for infrequent patterns
  • Fixed memory flush triggering too early on short conversations

v0.7

Hybrid Memory System

  • Memory
  • AI
  • Performance

A three-tier memory architecture that ensures Crew One always remembers what matters - your preferences, your history, and your commitments.

  • Tier 0 always-inject context (~500 tokens) in every request
  • Hindsight episodic memory with 4-network semantic search
  • Graphiti entity graph for relationship and commitment tracking
  • Cross-agent shared memory so agents learn from each other
Everything in this release
  • FastFacts store with categorised quick-access facts (profile, preference, contact, workflow, goal)
  • Token budget allocation per category to stay within limits
  • Hindsight 4-network architecture: world, experience, opinion, observation
  • Neo4j-backed entity graph for COMMITTED_TO relationship edges
  • Weekly, monthly, and quarterly memory consolidation
  • Shared memory provider for cross-agent context sharing
Fixes
  • Fixed Tier 0 exceeding token budget when user profile grew too large
  • Resolved Hindsight semantic search returning stale results after updates
  • Corrected Graphiti connection pooling under concurrent agent requests

v0.6

7-Agent Executive Assistant Team

  • Agents
  • Productivity
  • Feature

Meet your crew - seven specialised AI agents handling calendar, email, research, tasks, Slack, Notion, and file operations.

  • Calendar agent manages Google Calendar across multiple accounts
  • Email agent handles Gmail with smart categorisation
  • Research agent delivers web research and summarisation
  • Task agent tracks todos with priority and deadline management
Everything in this release
  • Calendar agent with multi-account Google Calendar support (business + personal)
  • Email agent with Gmail integration and smart filtering
  • Research agent for web search, summarisation, and deep dives
  • Task agent with priority levels and deadline tracking
  • Slack agent for workspace messaging and channel management
  • Notion agent for page creation, editing, and database queries
  • File agent for document operations and organisation
Fixes
  • Fixed calendar agent timezone handling for cross-timezone scheduling
  • Resolved email agent OAuth token refresh failures
  • Corrected task agent not respecting priority ordering in list views

v0.5

Web Dashboard & Chrome Extension

  • Frontend
  • UI
  • Extension

Access Crew One from your browser with a full-featured dashboard and a Chrome extension that brings your AI crew to any webpage.

  • Next.js web dashboard with Kanban boards and mission control
  • Chrome extension with side panel chat and command palette
  • BetterAuth integration for secure, cookie-based authentication
  • Real-time sync with optimistic updates via Zustand
Everything in this release
  • Kanban board with drag-and-drop, real-time collaborative editing
  • Mission control dashboard for agent activity overview
  • Chat interface with streaming AI responses via Vercel AI SDK
  • Chrome extension side panel for contextual AI assistance
  • Command palette (Cmd+K) for quick actions from any webpage
  • PostHog analytics integration for usage insights
  • Responsive design optimised for desktop and mobile
Fixes
  • Fixed hydration mismatch in server-rendered chat components
  • Resolved Kanban optimistic update conflicts during rapid edits
  • Corrected Chrome extension content script injection on SPAs
  • Fixed auth session expiry not redirecting to login modal