# Share one memory between several agents or tools for a person or team

> Share one memory between several agents or tools for a person or team: 6 companies fully do it, 15 with a limit, 55 carry the tag. Every listing in Agent memory with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/agent-memory.shared-memory/
- Markdown: https://www.anchorterminal.com/niches/agent-memory.shared-memory/index.md (~1,950 tokens)
- Slim: https://www.anchorterminal.com/niches/agent-memory.shared-memory/index.min.md (~430 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/agent-memory.shared-memory/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 6 | 15 | 55 | 8 | 6 |

## What counts as done

A fact written by one agent is readable by another in the same group within one search call, a fact in another user's or project's scope never appears, and a locked or canonical note can't be edited by an agent.

How agents ask for it: let Codex see what Claude Code learned; ten agents on one account graph; a team knowledge base our agents can all write to

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Amazon Bedrock AgentCore Memory (Amazon Web Services, Inc. (regional AWS entities by account location))**: Namespaces take custom variables such as org, team and actor, namespacePath reads a hierarchy, and IAM condition keys bedrock-agentcore:namespace restrict which principals read or write each namespace.
- **Basic Memory (Basic Memory LLC)**: Cloud team workspaces and team projects share one knowledge base, every call is project scoped, and locked: true makes a note read-only to all API and MCP writes, shipped after issue 993.
- **Graphicle (KarLex AI, Inc.)**: The team server gives each agent its own access key, revokes a key per project with graph.access, serves one project per HTTP endpoint, and agents can only propose memory, so reviewed memory stays canonical.
- **Honcho (Plastic Labs, Inc.)**: A workspace holds many peers and agents, peer- or session-scoped keys with an expiry limit what an agent can write, scopes bound recall, and the Claude Code plugin supports shared team workspaces.
- **Supermemory API + MCP (Supermemory)**: Each namespace has its own vector index so nothing leaks across, scoped API keys cannot cross a namespace, and MCP spaces give teams shared memory.
- **Zep (Zep Software, Inc.)**: Shared Context Graphs, UserGroups and ABAC policies on API keys govern who reads and writes which graph, and CallSphere ran ten specialist agents on one graph.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **AI Memory (Fabio Akita)**: Multi-user auth, per-project team sharing, a current-project pointer isolated per caller and claim-once handoffs are built in, but no lock flag was found to stop an agent editing a canonical page.
- **claude-mem (Alex Newman)**: TeamBrain adds team_brief and team_search over a hosted store and cloud sync carries one developer's database between machines, but pricing is unannounced and no lock or per-note permission exists.
- **Cognee (Topoteretes UG (haftungsbeschränkt))**: API mode shares one backend across MCP instances with dataset permissions, but recall takes dataset names, not ids, so a shared dataset another user owns can fail by name (documented warning).
- **Egoist AI Passport (Egoist Machines, Inc.)**: One owner's Passport is readable by several apps, each with its own passes, so a fact one assistant proposed reaches another only after the owner approves it; team sharing is not the design.
- **EverOS (Evermind AI, Inc.)**: Scopes by app_id, project_id, user_id and agent_id let several agents share a store, but the local server has no authentication and relies on binding to 127.0.0.1.
- **Agent Platform Memory Bank (Google LLC)**: Scopes can name any identity and IAM can restrict which principals read or write a scope's memories, but setup needs a Google Cloud project with billing created by a person, and no per-memory lock exists.
- **Graphiti (Zep Software, Inc.)**: group_ids isolate graphs and ten CallSphere agents shared one graph, but the reference MCP and REST servers have no authentication by design and are not for production.
- **Hindsight (Vectorize, Inc.)**: Banks are shared through agentBankMap and tags scope users within a bank, but an observation scope that leaves out user tags merges two users' facts into one observation, as the docs warn.
- **Hy Memory (Tencent)**: Several callers share a store through user_ids lists and source tags, but the bundled HTTP server has no authentication and allows any origin.
- **LangMem (LangChain (the licence file gives no legal form))**: Namespaces such as ("memories", "team_a") let agents write their own and read a shared team namespace in one deployment, but no auth or lock exists beyond the store you run.
- **Mem0 Platform + MCP (Embedchain, Inc. (DBA Mem0))**: user_id, agent_id, app_id and run_id scope memories and a search under one id never returns another's (one HN user hit this when onboarding under the wrong id); no lock or per-note permission exists.
- **MemMachine (MemVerge, Inc.)**: org_id and project_id isolate namespaces, but the self-hosted server accepts any caller, a shared group returns all members' data (issue 320), and 100 concurrent searches gave 30 per cent 500s on Neo4j, not planned to be fixed (issue 940).
- **Memori (Memori Labs Inc.)**: Attributes are shared per entity across processes and several agents can write under one entity, but there is no lock and the hosted key grants access to all entities.
- **MemPalace (MemPalace (open-source project))**: One hub with a writer lease serialises every agent's writes and remote agents use a bearer token, but it is one token with no scopes, read-only is server wide, and HNSW corruption with concurrent writers is an open issue in the map.
- **past.dev (Revo Inc.)**: Audiences, fixed or by trait rule, decide which identities read a data point and project keys scope access; but a re-send without audience resets the point to project visible, and nothing locks a record.
- **TencentDB Agent Memory (Tencent)**: Teams, private, team and restricted ACLs and agent loadouts share assets under human review in the panel, but the listing records Agent and Task reads unscoped until an unreleased 28 October commit.
- **Tokenrip (Quintel, LLC)**: Team workspaces with editor and viewer roles, guests, pinned decisions and a session log serve several agents, but a pasted tr_ key has full account access with no read-only key (map blocker).

- not this job: Context Mode Engine (MKSF LTD): The Engine is a local single-machine plugin; sharing across agents and machines is the Gateway's job.

## What stops agents finishing

- group search returns every member's data by design (MemMachine #320)
- self-hosted servers with no authentication (MemMachine, EverOS, Graphiti)
- one bearer token with no scopes or per-user keys (MemPalace, Tokenrip)
- HNSW index corruption with concurrent writers (MemPalace #1963)

## Around it

Before: agent-memory.coding-session-memory. After: agent-handoff.task-handoff.

## Where this comes from

Weight 8 of 100 in Agent memory: 6 of 36 accounts (MemMachine #320, Basic Memory #993, CallSphere's ten agents, two Mem0 HN commenters)

As data: https://www.anchorterminal.com/api/v1/use-cases/agent-memory.shared-memory.json, or the MCP tool get_use_case with id agent-memory.shared-memory. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
