Niches · Agent memory

Store a session and know it landed before relying on recall

Four or more companies' docs show a full fit.

7companies, full fit
14companies, fit with a limit
30companies carry the tag
10share of the category's work
8first-hand accounts

What counts as done

The write returns an id or event whose status the agent can poll to completed, a failed or truncated write returns an error rather than success, and a search after completion finds the fact.

Measured, for our own tests: the status reaches completed within 60 seconds. No docs we read state these figures, so they don't count against a listing here.

How agents ask for it: save this before I close the session, did that get remembered, flush memory now

Does the job 7

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

ListingWhy
AI Memory B
Fabio Akita
Writes are synchronous: memory_write_page returns a page id, the batch write commits all or nothing when a precondition fails, and memory_status reports counts. source
Basic Memory A
Basic Memory LLC
bm_write returns the path and checksum synchronously and notes show up in search moments after; refusals are MCP tool errors with codes such as REVISION_CONFLICT and ALREADY_EXISTS. source
Cognee D
Topoteretes UG (haftungsbeschränkt)
A blocking remember raises CognifyFailedError on failure, a background run exposes status running, completed or errored, and GET /api/v1/datasets/status reports indexing state. source
EverOS BB
Evermind AI, Inc.
add returns 200 with status extracted once the Markdown is on disk, or 202 with a request_id to poll at GET /tasks/{id} (queued, processing, success, failed with error_code); 413 and 429 refuse rather than drop. source
past.dev BB
Revo Inc.
A send returns 202 with an ingestionId, settlement is polled to completed or failed with a code, idempotencyKey makes retries safe, and recall marks pending ingestions; a large import can delay later sends. source
Supermemory API + MCP C
Supermemory
add returns an id and status queued, documents.get is polled to done or failed, and dreaming instant makes memories available as soon as the document finishes. source
Tokenrip B
Quintel, LLC
Writes are synchronous versioned artifacts that refuse without a why (WHY_REQUIRED) or on a stale revision (CONFLICT), and a missing scope fails with INSUFFICIENT_SCOPE. source

Does the job, with a limit 16

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.

ListingWhy
Amazon Bedrock AgentCore Memory A
Amazon Web Services, Inc. (regional AWS entities by account location)
CreateEvent is synchronous but extraction is asynchronous and the guide says a minute or more; the only checks are polling ListMemoryRecords and listing failed extraction jobs, with no per-write status. source
claude-mem C
Alex Newman
Hooks post observations fire-and-forget with a 2 second timeout and observation_generation_status reports the worker; issue 4068 shows 1,297 queued observations dropped on restart with no error to the agent. source
Context Mode Engine C
MKSF LTD
ctx_index writes synchronously to local SQLite, but the reference says nothing about errors on a failed or truncated index call. source
Agent Platform Memory Bank B
Google LLC
Generation is asynchronous and the overview calls it a background operation; the pages read do not show the status an agent polls or how long it takes. source
Graphicle BB
KarLex AI, Inc.
propose_memory is idempotent and returns a draft, and workflow_status lists queued jobs, but a write lands as a draft that is never served until a member approves it. source
Graphiti D
Zep Software, Inc.
The Python add_episode awaits extraction, but the reference servers queue episodes and answered 202 Accepted while nothing persisted, closed not planned (issue 566); get_status reports only the server and database. source
Hindsight D
Vectorize, Inc.
Synchronous retain returns success and items_count, async returns an operation_id to poll with a user-supplied id for safe retries; but the OpenClaw plugin stopped retaining after a reload with nothing logged until 1.0.0 (issue 4828). source
Honcho B
Plastic Labs, Inc.
Messages are stored at once and reasoning is queued with a queue status endpoint to poll, but the deriver was off and batches never flushed in 3.0.5 (issue 494) and a truncated peer card write returned success (issue 1143). source
Hy Memory C
Tencent
add() is synchronous and returns a result the caller checks, but MemHub users saw every write fail post-write verification after about 40 batches with a retry loop spending tokens (issue 72). source
LangMem D
LangChain (the licence file gives no legal form)
Store writes are synchronous and ainvoke returns the memories, but issue 154 shows nothing persisted with only a serialisation error in logs, and the listing notes open issues on success returned for failed writes. source
Mem0 Platform + MCP E
Embedchain, Inc. (DBA Mem0)
Platform add returns status PENDING with an event_id to poll at GET /v1/event/{id}; OSS add() returned normally after dropping memories on embedding failures, accepted in August and still open (issue 5245). source
MemMachine C
MemVerge, Inc.
Adding memories is a synchronous 200 or 422, but the docs do not say when semantic extraction finishes, and the listing records a provider 429 surfacing as a 500. source
Memori D
Memori Labs Inc.
Augmentation runs asynchronously after the conversation and the MCP tool stores after the agent drafts a reply, but no status or event id to poll was found. source
MemPalace B
MemPalace (open-source project)
add_drawer and mine are synchronous local writes, but status reported a wrong count after a large mine (issue 599), so the agent cannot trust it to know everything landed. source
TencentDB Agent Memory C
Tencent
session/add returns the object key and bytes synchronously with 400, 413, 422 and retryable 503 errors, but L1 to L3 extraction is asynchronous and the memory-generation-log routes were not read. source
Zep BB
Zep Software, Inc.
graph.episode.add returns the episode and a task that tracks its processing, and the listing records a 202 delay before facts appear; the pages read do not show the task status route or a time bound. source

Doesn't do this job 1

Listings in the category that do something else, with the reason.

ListingWhy
Egoist AI Passport C
Egoist Machines, Inc.
A write becomes a proposal the owner approves, so the agent cannot poll it to completed within 60 seconds. source

What stops agents finishing

  • 202 Accepted with no status route and nothing written (Graphiti #566)
  • deriver runs as a separate process that is off by default (Honcho #494)
  • plugin reload stops capture with nothing logged (Hindsight #4828)

Around it

Before: agent-memory.coding-session-memory, agent-memory.user-facts-recall. After: agent-memory.user-facts-recall.

Where this comes from

Weight 10 of 100 in Agent memory: 8 of 36 accounts are silent write failures (Mem0 #5245, Graphiti #566, Honcho #494 and #1143, Hindsight #4828, LangMem #154, Hy Memory #72); the commonest failure mode found

The same as data: /api/v1/use-cases/agent-memory.store-and-confirm.json, or the MCP tool get_use_case with id agent-memory.store-and-confirm. Other jobs in this category: /api/v1/niches?category=agent-memory.

Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.