LangMem

# Fix list: LangMem

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/langmem, the October 2026 research run, assessed 8 October 2026. Grade D, 47.9 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on LangMem: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 32 out of 100, up to 13.6 more on the total

Why it scored 32: Local-software reading. Installs from PyPI as `langmem` with Python 3.10 or later declared, but `extraction.py` on main uses `typing.NotRequired`, which arrived in Python 3.11, and issues 81 and 83 about failed imports on 3.10 are open. We read the source and did not run it on 3.10 (14 of 20). The repository holds tests (30 summarisation tests and a docstring-example runner), but its two workflows only deploy docs and publish to PyPI, so no public CI runs them (5 of 25). 54 open issues, 23 with no comment, among them a `TypeError` in summarisation from November 2025 (135), a schema mismatch between the tool and the store manager (138) and search missing memories written by the store manager (140) (8 of 25). Versions are 0.0.x with no changelog, GitHub releases or tags, and open issue 130 shows 0.0.30 was cut to restore compatibility with LangGraph 1.0 (5 of 15). Below 1.0 and not declared stable (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Security & auth, 45 out of 100, up to 9.6 more on the total

Why it scored 45: Local-software reading. The library holds no credential of its own and opens no port. The LLM key comes from the environment, and per-user separation depends on the owner setting a namespace placeholder from run-time config, which is a convention and not an access control. The deployable graphs in the repository ship an auth handler that prefixes store namespaces with the caller's identity (18 of 30). `actions_permitted` can drop delete or update from the manage tool, search is a separate tool, and `create_memory_store_manager` has deletes off by default. Nothing asks for confirmation before a delete (14 of 20). Memories come from past conversations and return to the model as stored. No guidance on poisoning or injection was found, and issues 163 and 164 proposing a guard (May 2026) have no reply (0 of 15). No audit log in the library. Store items carry created and updated timestamps (3 of 15). The organisation's SECURITY.md points to two Intigriti disclosure programmes, and an August 2026 commit resolves open Dependabot alerts. No `SECURITY.md` in this repository, no published advisories, no security.txt on langchain.com, and no certification was checked (10 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Maintenance & community, 15 out of 100, up to 7.4 more on the total

Why it scored 15: 0.0.30 reached PyPI on 27 October 2025, 346 days before the check (0). No release in the last 90 days (0). Of 54 open issues 23 have no comment, 19 pull requests are open with the oldest from June 2025, and nine of the ten pull requests opened since 1 August 2026 have no comment. The four issues filed between 5 and 8 October 2026 are too new to count against it (5 of 25). One Python package, current on PyPI only as the year-old 0.0.30, and no TypeScript package (5 of 15). Dependabot updates the lockfile, with the latest merge on 2 October 2026 and a pass over open alerts in August, but no workflow runs tests (5 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## 4. Schema & documentation, 55 out of 100, up to 7.3 more on the total

Why it scored 55: Library reading, with the tool lines applied to the two agent tools. No OpenAPI applies. Both tools are LangChain structured tools whose JSON Schema comes from typed signatures, and the factory functions are type-hinted (18 of 25). langchain-ai.github.io/langmem/llms.txt returns 404 (0). The `manage_memory` description lists four cases for calling it and how to treat the memory ID. `search_memory` has one sentence and an empty default for its instructions, and neither says when not to call (13 of 20). `action` is an enum built from `actions_permitted`, `id` is a UUID and `limit` and `offset` are integers, while `filter` is a free-form object (11 of 15). Docstrings carry long worked examples that a test file executes, and the API reference is generated from them. Error behaviour is not documented (10 of 15). Version numbers exist only on PyPI, with no changelog (3 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 5. Agent ergonomics, 66 out of 100, up to 5.5 more on the total

Why it scored 66: Two tools with short descriptions (25). `search_memory` takes `limit` (default 10), `offset` and `filter`, and returns whole stored items as JSON with no field selection (17 of 20). Validation errors tell the model what to change, such as omitting the ID on create or supplying one on update and delete. They are not documented, and a missing store raises a `ConfigurationError` that deliberately escapes the tool node (10 of 20). Update and delete by ID are safe to repeat. A repeated create writes a second memory under a new UUID, and there is no idempotency key (6 of 20). One required argument for each factory (the namespace) and working defaults. Python only, with an open request for TypeScript (issue 121) (8 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 6. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: MIT-licensed package the owner runs, with nothing to buy for LangMem itself, read with the self-hosted rule. 20 for pricing, 20 for a free start and 20 for use without an account. No x402, MPP or L402 in the docs or source (0). LLM, embedding and database costs are the owner's.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 7. Transparency & trust, 59 out of 100, up to 3.6 more on the total

Made of editorial 63, provenance 55.

Why it scored 59: MIT licence in the repository and on PyPI (30). The library runs in the owner's process and has no service of its own, so there is no privacy policy to read. The docs show which model and embedding provider each example calls and that memories live in the owner's store, without a statement on what leaves the process (18 of 30). No deprecation policy or dated notices were found (0 of 20). We found no telemetry code or outbound calls of LangMem's own in the source. Prompt optimisation wraps its steps in `langsmith` trace blocks, and the docs do not say when those send data (15 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: langchain.com, registered 2019-12-03 (6 years) (11 of 15)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- We read `typing.NotRequired` in the source and the two open issues but did not run the package on Python 3.10.
- Maintainer response is judged from comment counts on open issues and one thread read in full. Closed issues were not sampled.
- unchecked: whether `langsmith` trace blocks in the prompt optimisers send anything without tracing environment variables set.
- unchecked: LangChain's certifications and the terms of its hosted products, which do not govern this library.
- Whether LangChain plans further LangMem releases is not stated in the repository or docs.
- Issues 196 and 198, filed on 8 October 2026, report that the store manager returns success when a write fails and ignores `query_limit` above the store default. Neither was verified.

## Weaknesses

- The newest PyPI release, 0.0.30, is from 27 October 2025, and every commit to `src` on main is older than that
- No changelog, GitHub releases or tags, and versions are still 0.0.x
- The repository's two workflows deploy docs and publish to PyPI. Neither runs the tests
- 54 open issues, 23 with no comment, and 19 open pull requests, the oldest from June 2025
- No guidance on memory poisoning or prompt injection found, and two open issues asking for it (May 2026) have no reply

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Pass a store with an embedding index (`index={"dims": 1536, "embed": "openai:text-embedding-3-small"}`), or `search_memory` cannot rank by meaning
- Use a database-backed store such as `AsyncPostgresStore` for anything that must survive a restart. `InMemoryStore` loses everything
- Put a per-user placeholder in the namespace and set it in `config["configurable"]` on every call, or users share one memory space
- Send the memory `id` with `update` and `delete`, and never with `create`. A retried `create` writes a duplicate under a new UUID
- Run on Python 3.11 or later. `src/langmem/knowledge/extraction.py` uses `typing.NotRequired`, which Python 3.10 lacks

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.