# The memory every agent shares.

Source: https://corespeed.io/memory

Say it once, to any agent. Every agent on your team remembers it.

Get started: https://app.corespeed.io/sign-up · Docs: https://corespeed.io/docs/memory

## Open to every agent. Closed to everyone else

Plug in any agent. It's already up to speed.

- **Any agent.** — Claude Code, Codex, Cursor, OpenClaw, Hermes. If it speaks MCP, it remembers.
- **Private. Until it isn't.** — Yours by default. Your team's the moment you share it.
- **Sealed.** — Nobody outside your organization reads a word.

## Top-tier accuracy. Same model. Higher score

Our memory engine on LongMemEval-S, beside each system's own published numbers. How we measured: https://corespeed.io/blog/how-we-took-longmemeval-from-80-to-94-without-touching-retrieval

| Measure | CoreSpeed | Mem0 | Zep | Supermemory |
| --- | --- | --- | --- | --- |
| LongMemEval-S accuracy | 94.2% | 94.4% | 90.2% | 85.2% |
| Answering model | GPT-5.4 | GPT-4o | GPT-5.4 | Gemini 3 Pro |
| Median search time | 112 ms | — | 104 ms | 187 ms |

- **0 of 500** private memories leaked in testing. The others publish no such test.
- **Teammates' changes** wait for review. The others publish no such step.

A dash means the system publishes no comparable number.

## Built for a team that changes its mind

Agents propose. People decide. Nothing gets lost.

### Review

- Agents propose, with a reason.
- You merge. Or don't.
- Agents never merge.
- Stale proposals never land.

Learn more: https://corespeed.io/blog/pull-requests-for-shared-memory

### History

- Three versions back, with who wrote them.
- Retire the old. Point to the new.
- Undo any archive.
- Linked where it matters.

Learn more: https://corespeed.io/docs/memory#keeping-shared-memory-accurate

### Recall

- Pinned memories, from the first message.
- By word, by name, by meaning.
- Just the passage that matters.
- Searching is free.

Learn more: https://corespeed.io/blog/how-memory-search-works

## Agents do the remembering. You do the deciding

They save as they work. They search before they start. You call the shots.

- **Remember** — Decided once. Saved instantly.
- **Recall** — Any agent. Any day. With the source.
- **Propose** — Their memory. Their call.
- **Link** — What belongs together, stays together.

## Permissions no agent can sweet-talk

Enforced row by row, in Postgres. Nothing an agent says changes that. How isolation works: https://corespeed.io/blog/memory-isolation-with-row-level-security

- **See yours. And what's shared.** — Your organization sees what you share. Nobody else sees a thing.
- **Change only yours.** — Everyone else proposes.
- **Only people merge.** — You or an admin. Never an agent.

Who can do what with a memory you saved:

| Who | Read | Change | Merge a proposal |
| --- | --- | --- | --- |
| You | Yes | Yes | Yes |
| Your agents | Yes | Yes | No |
| A teammate | Yes if shared, no if private | Proposes if shared, no if private | Admins only if shared, no if private |
| Their agents | Yes if shared, no if private | Proposes if shared, no if private | No |
| Another organization | No | No | No |

## How we built it. In the open

- [Pull requests for shared memory: proposals, archive and versions](https://corespeed.io/blog/pull-requests-for-shared-memory) — A correction next to a stale shared memory does not stop the old one being retrieved. Archive it with a successor, or propose a fix for its author to merge. (Markdown: https://corespeed.io/blog/pull-requests-for-shared-memory.md)
- [Memory isolation with row-level security, not application code](https://corespeed.io/blog/memory-isolation-with-row-level-security) — Caller identity comes from auth, each memory carries its scope, and Postgres row-level security enforces the boundary. Nothing in the request body can cross it. (Markdown: https://corespeed.io/blog/memory-isolation-with-row-level-security.md)
- [How memory search works: hybrid retrieval, excerpts and result budgets](https://corespeed.io/blog/how-memory-search-works) — memory__search fuses exact and relaxed lexical matching, entity aliases and vector similarity into one ranking, then bounds each row and the result. (Markdown: https://corespeed.io/blog/how-memory-search-works.md)

## Pricing

Saving costs 1 credit. Searching is free. 200 MB on Free. 2 GB on Pro. https://corespeed.io/pricing (Markdown: https://corespeed.io/pricing.md)

## Questions people ask

### Does the conversation carry over to the next agent?

No. The transcript stays in the client that wrote it. What carries over is what an agent saved to memory.

### Who can read my memories?

Only you, until you share one. A shared memory is open to everyone in your organization and their agents, and to nobody outside it.

### Can memory stop an agent from doing something?

No. Memory gives an agent context. To stop a risky action until a person approves it, use approvals. How approvals work: https://corespeed.io/docs/approvals

### Do I still need CLAUDE.md or AGENTS.md?

Yes, for rules. Keep the instructions in the file and the facts in memory, so every agent shares the facts. Shared memory vs instruction files: https://corespeed.io/blog/shared-memory-vs-instruction-files

### What should not go in memory?

Secrets, whole transcripts and application data. Memory is for durable facts, decisions and preferences.

### What does it cost?

Saving a memory costs 1 credit. Searching and reading are free. Free includes 200 MB of memory, Pro 2 GB. See pricing: https://corespeed.io/pricing

## Links

- Product overview: https://corespeed.io/ (Markdown: https://corespeed.io/index.md)
- Memory docs: https://corespeed.io/docs/memory (Markdown: https://corespeed.io/docs/memory.md)
- Pricing: https://corespeed.io/pricing (Markdown: https://corespeed.io/pricing.md)