Skip to content
Use cases

For agents that should stop forgetting

Coding agents that hold your conventions, research agents that keep their findings, support agents that recall the last fix — and teams that share one memory instead of starting cold every session. Connected over MCP, visible and editable throughout.

Coding agents

Conventions, decisions, and gotchas — remembered

Claude Code and Cursor stop re-learning your codebase every session. The choices you made, the patterns you follow, and the traps you already hit become memory they can read on the next task.

  • Coding conventions and project structure your agent should follow
  • Architectural decisions and the reasoning behind them
  • Known gotchas and fixes, so the same bug isn't re-debugged twice
claude-code · shared memory
claude-coderemembers

We use Postgres 16 in prod, and all timestamps are stored in UTC. Never run migrations directly against prod.

later, a new session

asks"how should I add the new audit table?"

recalls the UTC-timestamp convention and the rule to ship migrations through CI — not straight to prod.

Sourced, visible, and editable — the same shared memory every connected agent can reach.

Research agents

Findings that don't evaporate

A research agent gathers sources, extracts findings, and forms a picture over many runs. Hold that picture in a graph so the next session builds on the last instead of starting from a blank page.

  • Sources and the findings drawn from them, kept across runs
  • Entities and how they connect, growing into a graph over time
  • Provenance on every memory, so a claim traces back to where it came from
research-agent · shared memory
research-agentremembers

Library A and Library B both parse the format, but only A streams large files; B loads everything into memory.

later, a new session

asks"which library should we use for multi-GB inputs?"

recalls the comparison it built earlier — A streams, B doesn't — without re-reading every source.

Sourced, visible, and editable — the same shared memory every connected agent can reach.

Support & ops agents

The last fix, on hand

Support and ops agents see the same kinds of incidents again and again. Keep the resolutions, the workarounds, and the customer context as memory, so the next occurrence starts from what already worked.

  • Past incidents and the resolutions that actually fixed them
  • Customer and account context that should carry between conversations
  • Runbooks and workarounds, editable the moment they change
ops-agent · shared memory
ops-agentremembers

Error E-4012 on Acme's account is caused by an expired webhook secret; rotating it clears the alert.

later, a new session

asks"Acme is seeing E-4012 again — what do we do?"

recalls the known cause and the fix — rotate the webhook secret — instead of triaging from scratch.

Sourced, visible, and editable — the same shared memory every connected agent can reach.

Teams

One memory, shared by the whole team

When everyone's agents write to the same workspace, what one person's agent learns is there for the next. Roles keep it governed, and per-org API keys connect each agent to the same graph.

  • One shared memory across the org — not a SQLite file stuck on a laptop
  • Owner, admin, and member roles over that single graph
  • Per-org API keys connect every agent to the same memory
team workspace · acme-research
alex's cursorremembers

The staging deploy key lives in 1Password under 'infra · staging', and deploys go out via the release channel.

later, a new session

asks"jordan's claude-code: how do I deploy to staging?"

recalls the process Alex's agent recorded earlier — same workspace, same shared memory, governed by role.

Sourced, visible, and editable — the same shared memory every connected agent can reach.

Give your agents a memory to share

Connect Claude Code, Cursor, or any MCP client on the Free plan, and watch the shared memory start to compound.

No credit card required · Free plan available · Bring your own OpenAI key

SEE WHAT YOUR AGENTS REMEMBER