Now in beta

Stop re-explaining
your company to every AI.

Every workspace ships with @workspace — an agent that knows your company because it shares memory with your team. Plug in Cursor, Claude Code, ChatGPT and your own bots through MCP, and they all read from the same living source of truth.

alice / Personal drive/decisions/2026-q3-roadmap.md
2026-q3-roadmap.md×
auth0-rollback.md
# Q3 roadmap decisions
captured from#engineeringPR #482ENG-1244

## What we're shipping

- Visual flows for Slack → agent → drive.

- GitHub PR triggers w/ auto-summarized diffs.

- Jira wire-up (q3.5).

## Open threads

-Live cursor on long docs?
-@bob is typing…
Read & written by your agents
  • Cursor
  • Claude
  • ChatGPT
  • VS Code
  • OpenAI Agents SDK
  • LangChain
  • Vercel AI SDK
  • Vertex AI
  • Copilot Studio
  • GeminiSoon
  • JetBrains AISoon
Integrates with
  • Slack
  • GitHub
  • Drive
  • Webhooks
  • Schedules
  • JiraSoon
  • LinearSoon
  • NotionSoon
5 agents · 1 document
Your workspace agent

Meet @workspace — an agent that already knows your company.

Every workspace ships with a Claude-powered agent that reads your drive, greps captured Slack and GitHub events, follows concept tags, and answers with citations. No prompt engineering — just mention @workspace and it shows up with the receipts.

  • Answers grounded in your drive

    Searches files, summaries, and captured events. Every answer carries citations back to the Slack thread, PR, or doc it came from.

  • Writes decisions back to shared memory

    Promotes a thread into a versioned markdown decision in one click — so the next agent that needs the answer just reads it.

  • Mention it anywhere

    @workspace works inside any chat or file. Your team and your agents converse with the same memory.

MCP · the agent wedge

Plug every other agent into the same living memory.

Cursor, Claude Code, ChatGPT, VS Code, and your own bots all authenticate once, then read and write against the same source of truth — search files, follow concepts, drop summaries, file decisions back. No glue code, no prompt dumps, no copy-paste.

Cursor

Editor + agent mode

Claude

Code & Desktop

ChatGPT

Custom MCP app

VS Code

Copilot agent mode

OpenAI Agents SDK

hostedMcpTool

LangChain / LangGraph

MCP adapter

Vercel AI SDK

createMCPClient

Vertex AI

Agent Builder

Copilot Studio

Admin-gated MCP

GeminiSoon

Enterprise Agent Platform

JetBrains AISoon

IntelliJ / WebStorm

The bones

Built on standards your team already trusts.

Protocol
MCP-native, OAuth 2.1 + DCR
Built-in agent
Claude tool-use over your drive, with citations
Concepts
AI summary + topic tags per file, drive-wide rollup
Collaboration
CRDT-backed, millisecond conflict-free
Search
BM25 + grep across shared memory
Storage
Markdown on disk, Git-friendly diffs
Auth
OAuth 2.1 + OIDC, PATs for scripts
The substrate

An agent is only as good as the memory underneath it.

Today that memory is scattered across Slack threads, PR comments, CRM notes, call recordings, stale Notion pages, and prompts pasted into five different AI tools. And your agents keep producing more of it — PRDs, decision logs, call summaries, planning docs, design notes — all in markdown. Where does any of it go?

is where. One shared layer — readable, editable, and versioned — so @workspace, your coding agents, and the humans on your team all reason from the same ground truth.

  • Markdown you can trust forever

    Drives, files, chats. Versioned, diffable, search-indexed. Portable by default — nothing sits in a vanishing chat window or a vendor black box.

  • Live chat & live docs

    Every conversation and every document is shared in real time. Humans and agents reply, edit, and review in the same thread — every cursor, every change, visible the instant it happens.

  • Concepts: your drive as a knowledge graph

    Every file gets an AI summary and weighted topic tags. ctxhub rolls them up into a ranked map of concepts per drive — click any concept to see which files discuss it, and let agents follow the same trail.

Drive1,247notes18contributors
Flows · NEW

Intelligent automation
that thinks for itself.

Flows turn shared memory into agentic automation. Stand up a persistent agent that watches the events you care about — a PR opens, a deal moves, a thread heats up — pulls the right context from shared memory, reasons, and acts. Triage, summarize, draft, escalate, reply. The work gets done before you think to ask.

  • React to changes across your whole org

    A PR opens. A deal moves stage. A Slack thread heats up. A customer hits a milestone. A weekday tick. Any of them can wake the agent — without anyone in the loop.

  • Wire in the agent without writing code

    Stand up an agent workflow in an afternoon instead of a sprint. No webhook servers to host, no queues to babysit, no schedulers to wire — just the graph and the agent doing the work.

  • Send back to the tools you already use

    Reply on Slack, post to a channel, record a decision in shared memory, hit a webhook. One canvas spans the whole loop.

  • Build for the tools we don't ship

    Flows reach beyond what we ship — your CRM, internal APIs, custom AI services, and any enterprise system you can build against. No waiting on us to ship the integration you need.

Flow editor
#Slack messagethenmsgOnly ifinthatthenAAsk the agentinpromptthensave asReply on Slackintextmessage channel = engineeringmessage text/flows/triage-eng-decisions.flowLIVE
Sources & destinations

Plug into the stack you already run on.

Slack

Trigger on messages, post replies, ping channels.

GitHub

PR-opened triggers, file & decision links.

Drive

File-saved triggers feed back into the loop.

CRON / Schedules

Cron-style timers; weekday windows.

Webhooks

Outbound HTTP to anything you can hit.

Jira
Soon

Issue events, comment & status writes.

Linear
Soon

Issue & cycle triggers, status writes.

Notion
Soon

Page edits, database row writes.

Salesforce
Soon

Opportunity & account events; write call notes back.

HubSpot
Soon

Deal-stage triggers; sync meeting summaries.

Attio
Soon

Workspace events; enrich records from shared memory.

Attention
Soon

Call insights in; AI follow-ups out.

How it works

From scattered scraps to shared memory in three moves.

alice / decisions
  • 2026-q3-roadmap.md
  • from-slack.mdslack
  • auth0-rollback.md
  • kickoff-notes.md
1. Capture

Wire ctxhub into the stack your team already runs on — Slack, GitHub, Salesforce, HubSpot, plus anything else you build in Flows. Capture begins the moment you connect.

Cursor · MCPlive
@workspace what changed about Q3?
@workspace
3 decisions changed:
  • • Live cursor → q3.5
  • • Jira slipped
#engPR #482
2. Connect

Point any MCP-aware client at ctxhub — Cursor, Claude Code, ChatGPT, your own bots — and they all read the same source of truth. No glue code, no prompt dumps.

#q3-launch · liveAWB3 here
aliceshipping Visual Slack flows ✓
@workspacecaptured #eng + PR #482
bobtyping…
3. Collaborate

Your team and your agents reply in the same chat threads and edit the same documents — live, with every cursor, edit, and reply visible the instant it happens.

Give every agent on your team
the same memory.

Start with @workspace. Plug Cursor, Claude Code, ChatGPT, and your own bots in through MCP. Stop teaching every new AI from scratch — build a body of context that compounds with every conversation.