Your agentic team’s memory. With “governance” and “provenance”
Nothing reaches the shared graph without review, and every memory keeps its receipts — who saved it, where it came from, when it happened, and every edit since. Your agents get the full context, not a lossy summary.
Which edge cases broke last time?What did customers say on the last call?Give the new hire the real onboarding guide.Which edge cases broke last time?What did customers say on the last call?Give the new hire the real onboarding guide.Which edge cases broke last time?What did customers say on the last call?Give the new hire the real onboarding guide.Which edge cases broke last time?What did customers say on the last call?Give the new hire the real onboarding guide.
Find the decision behind this code.Reconnect the incident to the lesson.What should we never repeat?Find the decision behind this code.Reconnect the incident to the lesson.What should we never repeat?Find the decision behind this code.Reconnect the incident to the lesson.What should we never repeat?Find the decision behind this code.Reconnect the incident to the lesson.What should we never repeat?
Why does checkout retry twice?Where is the source of truth?What changed since the last release?Why does checkout retry twice?Where is the source of truth?What changed since the last release?Why does checkout retry twice?Where is the source of truth?What changed since the last release?Why does checkout retry twice?Where is the source of truth?What changed since the last release?
The Cortex platform
A little less repetition. A lot more collective intelligence.
Your tools create the context. Cortex turns it into knowledge your people and agents can build on.
Illustrative workflow
acme / payments main
01async function retryPayment() {02const attempts = 203// The why belongs with the code.04 }
Retry boundary → provider timeoutImplementation linked to the decision
GitHub + code memory
Your codebase. With a memory.
Connect a repository to a project. The CocoIndex-powered indexer turns code into searchable memories, so your next agent starts with the implementation in context.
Code indexing requires the connected indexer service.
Illustrative workflow
#engineering Source connected
“Let’s cap retries at two. The provider times out after the third attempt.”
Bind a connected source to a project. New activity becomes draft memories, ready for owner review or your configured auto-merge policy.
Illustrative workflow
decisionimplementationlesson
Connections found. Context restored.
Dreams
Wake up to connected knowledge.
Schedule a review of shared memories. Dreams pulls out the key facts from each one, updates useful links, and flags possible duplicates for you to review.
Illustrative workflow
Your week in context Weekly briefing
01 The retry policy changed
02 A deployment lesson worth keeping
03 What the team learned next
Less catching up. More moving forward.
Digests
Catch up without the archaeology.
Get a weekly digest of what changed and what your agents learned. Decisions, discoveries, and project updates in one readable briefing.
Illustrative workflow
Why do we retry twice?
The provider times out on attempt three. The team capped retries to avoid duplicate charges.↗ Payment retry policy
The answer. And the reasoning behind it.
Chat
Ask the team. Even after hours.
Ask Cortex about your memories and connected tools. Follow the sources, recover the reasoning, and carry the answer into your next task.
Illustrative workflow
PROJECT ACCESS BOUNDARY
Decision
Code context
Project members only Access scoped
Your team sets the boundaries.
Private draftOwner review Shared
Governance + provenance
Shared knowledge. Earned trust.
Keep drafts private until review. Every memory keeps its author, source, and edit history. Approve what belongs, explore how it connects, and keep each project inside its own access boundaries.
Integrations
Different tools. Shared understanding.
The decision in Slack. The change in GitHub. The thinking in your docs. Bring it all into orbit around your team.
You choose the accounts. Project membership controls access to shared memories.
02 / From lost to connected
Someone leaves. The knowledge stays.
It's 2 a.m. An incident hits the system your teammate built. Cortex gives you a place to find their reasoning, the known edge cases, and the decisions behind the code.
GitHub, Slack, Gmail, Notion, Granola, Google Docs
Email, calendar & transcript recipes
GitHub, Gmail, Notion, Linear, Granola & more
Agent sessions, Slack, GitHub, tickets, calls
Slack, Gmail, GitHub & more
Email, calendar, Slack, Drive, local files
API & integrations
Merged PRs → the “why” behind the code
PR-why + gotcha memories¹
—
—
PRs ingested as context
—
—
—
Background memory upkeep
Dreams: facts, links, duplicates
Dream cycle
Dynamic Dreaming
—
Self-updating wiki
Ongoing brain upkeep
—
Source attribution
Author, source, date, edit history
—
—
—
Wiki entries link to sources
—
—
Review before it becomes team knowledge
Draft → merge request → owner or policy approval
—
—
—
—
—
—
Chat with your work context
Built-in chat + Slack bot
—
Company Brain in Slack
Web app & Slack
App, Slack & iMessage
Built-in workspace
—
Weekly team memory briefing
Digests
—
—
—
—
—
—
Runs on your own infrastructure
Enterprise single-tenant
Self-hosted by default
Self-hosted server
—
—
Local on your Mac
OpenMemory (local)
Based on linked public product pages, docs and changelogs, reviewed October 28, 2026. A dash means this specific workflow was not documented, not that the product cannot support it. Features and plan availability vary. ¹ Requires connecting a GitHub repo to the code indexer; distilled from merged PRs, up to 150 code memories per org per day.
Pricing
Flat per seat, no surprise invoice.
One workspace for your team's context. Start free, add your team when you are ready. Hosted infrastructure is taken care of.
Free
$0forever
One person, your own LLM key. The full memory layer, with Claude Code hooks included.
A governed memory layer for your team and its AI agents. It stores decisions, lessons, and context as linked memories — each with its author, source, and edit history — and serves them back over MCP and a web UI: searchable, organized by project, and shared across the whole org.
How is this different from a vector database or my own RAG?
A vector DB gives you similarity search over chunks. Cortex gives you a curated team memory: structured entries with authorship, source, confidence, edit history, and links between related facts — plus hybrid full-text + semantic search, draft/approval workflow, and MCP tools your agents already know how to use. You could build it; this is it built.
How is this different from GBrain or Supermemory?
GBrain is one person’s brain: markdown files and a database on your own machine, maintained for your own agents. Supermemory is a memory API that now also offers a Slack-based Company Brain. Cortex is built for a team’s coding agents: memories are scoped to projects, new knowledge waits as a draft until an owner or policy approves it, every memory keeps who saved it and where it came from, and merged pull requests are distilled into the reasoning behind the code.
What do you mean by governance and provenance?
Governance: on projects with approval turned on, new memories land as private drafts and only reach the shared graph once a project owner (or your auto-merge policy) approves them. Provenance: every memory records who saved it, the source it came from (a PR, a Slack thread, a meeting, a doc), when the underlying thing happened, and its full edit history — so an agent can tell you not just what the team decided, but where that came from.
Who can see what?
Memories are scoped to projects inside your org, and project members only see their projects. On projects with approval turned on, a member’s saves stay private drafts until an owner approves the merge request — so the shared graph only holds reviewed knowledge.
Which tools does it work with?
Anything that speaks MCP: Claude Code, Cursor, Claude Desktop, custom agents. MCP tools cover saving, searching, context assembly, graph traversal, merge review, and searching your connected Gmail, GitHub, Notion, Slack, and Granola. A single search returns full memory content plus Dreams-extracted key facts, so agents answer in one call. There’s also a built-in chat, a full REST API, and a web UI.
Where does my data live?
In Postgres, on infrastructure we run and back up. Every memory belongs to one org and every read is scoped to the projects you’re a member of. Need your own isolated instance instead? The Enterprise plan gives your org a dedicated, single-tenant deployment. Fully on-prem and air-gapped installs are on our roadmap.
Is Cortex open source?
Cortex is a hosted product. A single-tenant deployment — a dedicated instance, isolated from other customers — is available on the Enterprise plan for teams that need it.
cortex