Roadmap

What's built, what's next, what's coming after.

Each milestone is a branch off the spine. Click a card to expand its scope. Shipped, in progress, planned, colour-coded so you can see where we are at a glance. We're building in the open and we'd rather show the honest state than over-promise.

● shipped ● in progress ○ planned
shipped

M0 · Foundations

2026

The base everything else stands on: multi-tenant chat workspace, connector framework, raw ingestion.

+ expand details
  • Multi-tenant workspace model, channels, threads, direct messages, membership
  • Connector framework: a single contract every integration follows
  • Bronze ingestion layer, raw artifacts land safely and idempotently
  • Fathom connector (meetings) and Gmail connector (mail) wired end-to-end
  • Usual chat application functionality, the surface your team already understands
in progress

M1 · Silver layer, the Cortex

Now

The smart context map under everything. A structured LLM-aware wiki that connects information in a way agents can actually use.

+ expand details
  • Normalise every source, meeting, email, document, ticket, into one queryable shape
  • Closed entity types so the structure stays trustworthy
  • Smart cross-references between people, projects, decisions, and the artifacts that mention them
  • Files are the source of truth; the database is a derived index that can be rebuilt
  • Anti-hallucination by construction, bodies stay verbatim, only navigation fields are LLM-filled
planned

M2 · Retrieval chat agent

Next

The main feature. Mention an agent in any channel. Ask a question. Get an answer grounded in your team's work, with sources.

+ expand details
  • @-mention any agent in any channel and ask a real question
  • Answers cite the meetings, emails, documents, and tickets they came from
  • Agent respects channel permissions, it only sees what the channel can see
  • Streaming responses inside the conversation, not in a separate window
planned

M3 · Google Drive, documents

Then

Bring your team's documents into the Cortex so the retrieval agent can answer questions about what's in them.

+ expand details
  • Hybrid Google Picker + custom folder browser for selecting what's in scope
  • Coverage for the file types that actually matter: docs, sheets, slides, PDFs
  • Shared Drives included from day one
  • Progressive OAuth scope, start narrow, upgrade only when you need more
planned

M4 · Draft & generation agent

Then

Generate drafts and documents inside the chat session. Anybody can see them, read them, and modify them in the same thread.

+ expand details
  • Co-write a document with an agent inside the channel where the conversation lives
  • The draft is visible to everyone in the channel and editable in place
  • Grounded on the Cortex layer, drafts cite the work they're built on
  • Versioning so you can see how the doc evolved
planned

M5 · WhatsApp, chat ingestion

Then

Bring WhatsApp conversations into the Cortex so client chats stop disappearing into someone's personal phone.

+ expand details
  • Read-side integration, chats land in the workspace's knowledge layer
  • Per-chat opt-in so personal conversations stay personal
  • Cross-user deduplication so the same thread isn't ingested five times
  • Mitigations around ban-risk and identity boundaries
planned

M6 · Linear & GitHub

Then

Close the loop with the tools where delivery actually happens, tickets, pull requests, releases.

+ expand details
  • Linear: tickets, comments, state transitions, project context
  • GitHub: issues, pull requests, reviews, releases
  • So the agent can answer "what's the status of the Acme integration?" without you opening four tabs
  • For tech-aligned teams: the substrate that helps engineering ship the right thing
planned

M7 · Private DMs with agents

Then

Any member of the company can DM an agent privately, exactly like messaging a colleague.

+ expand details
  • One-on-one conversations with any agent, in a private DM thread
  • Agent keeps a per-user memory, your style, your projects, your preferences
  • For sensitive prep, drafting, decisions you don't want a whole channel watching
  • Same access rules as a human colleague, the agent only sees what you'd let a teammate see
planned

M8 · Learning personalities

After 1.0

The small but magical detail. Agents that can read the room and jump into a conversation when it's useful, to calm a thread, clarify confusion, or share what they know.

+ expand details
  • Agents learn the people they work with, tone, context, what each person cares about
  • An agent can voluntarily step into a multi-person conversation when there's signal it can help
  • Defuse confusion: pull up the actual decision when two people remember it differently
  • Tailor responses per recipient, the same answer, framed for the person reading it
  • The detail that turns "an AI in the chat" into "a colleague"