Leon

# Fix list: Leon

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/leon, the October 2026 research run, assessed 12 October 2026. Grade E, 41 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 Leon: 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: Read with the local-software lines, as software the owner runs. The graded edition is the 2.0 Developer Preview on `develop`, which the README names as the project's focus. It installs by cloning the repository and running pnpm, with Node.js 24 or later and Linux, macOS and Windows stated, and there is no package or tagged build of it. The npm CLI (`@leon-ai/cli` 1.5.0 of 13 May 2023) installs the legacy line (10 of 20). The Tests, Lint and Build workflows trigger only on `master`. On `develop` a scheduled fresh-install workflow that installs Leon and runs DeepSeek inferences passed on every run listed from 27 July to 12 October 2026, and the 63 spec files in the repository are not run by CI on that branch (10 of 25). 114 open issues and pull requests, with setup failures reported in 2024 and 2025 (#534, #542, #546, #555) carrying no reply (8 of 25). `CHANGELOG.md` has breaking-change sections but ends at 1.0.0-beta.8 of 1 May 2023, and the preview has no version of its own (4 of 15). Beta and Developer Preview (0 of 15).

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, 34 out of 100, up to 11.6 more on the total

Why it scored 34: Harness reading of the framework checklist, as the Goose dossier used. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. Telemetry is on in the sample config and posts an instance ID, versions and hardware details to telemetry.getleon.ai, the install step registers the instance ID before any setting is read, the server listens on 0.0.0.0 and client authentication is off by default. Against that, connected-app tokens are encrypted with AES-256-GCM per profile, though the key sits in the same profile's `.env` beside plain-text model keys (10 of 30). No approval prompt before a tool call was found. Tool allow and disable lists, a read-only Gmail scope, a shell blocklist and a visible terminal for `sudo` are the controls, there is no sandbox, and the proactive pulse is on by default (6 of 20). The Gmail and web-fetch tool guidance says fetched content is untrusted data and not instructions, and the persona prompt says to ask first when an action is externally visible, both of which are prompt text and not code checks (5 of 15). Per-profile conversation logs, tool-call logs with inputs and output previews, and token usage logs (11 of 15). No `SECURITY.md`, no security.txt (404), no bounty and no published advisory. A code of conduct and an issue tracker exist (2 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. Schema & documentation, 33 out of 100, up to 10.9 more on the total

Why it scored 33: Harness reading of the framework line, as the Goose dossier used. Each of the 48 tools has a `tool.json` with JSON Schema parameters for its functions, 228 in all, and the profile config names a JSON schema. No OpenAPI file or written contract for the HTTP API or the Socket.IO interface was found (12 of 25). No llms.txt on getleon.ai (404) (0 of 10). The README says the 2.0 documentation is not ready, docs.getleon.ai describes 1.0.0-beta.7, and the current references are two generated context files, with tool guidance that says when to prefer each tool (6 of 20). Tool schemas use required fields, bounds and closed objects in places (Gmail `maxResults` 1 to 100), while the memory `write` options take free-form strings and an open metadata object (9 of 15). A commented sample config, with no worked examples or error reference for 2.0 (3 of 15). The changelog stops in 2023 and the preview is unversioned (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.

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

Why it scored 15: The last tagged release of the core is 1.0.0-beta.8 of 1 May 2023. The newest entries on the releases page are pre-release builds of the TCP server and the two bridges from 19 February 2026, 235 days before this check, so either reading is older than 180 days (0 of 30). No release in the last 90 days (0 of 20). The author committed at least 200 times between 25 September and 12 October 2026, but outside pull requests from March to September 2026 (#558, #565, #582, #589, #591) and most open bug reports have no comment (6 of 25). Not an MCP server, no packaged build of the preview, and the npm CLI was last published in 2023 (2 of 15). Dependencies are pinned at current versions, dependency graph updates run, and the fresh-install check passes on `develop` (7 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.

## 5. Agent ergonomics, 57 out of 100, up to 7 more on the total

Why it scored 57: Harness reading, scored on what a person or pipeline driving it has to supply. Toolkit schemas load only when needed (`progressive_toolkit_loading: true`), large tool outputs are kept as artifacts with bounded previews, and older exchanges are compacted (20 of 25). `agent_max_iterations` defaults to 256 and can be set per profile, with a fixed input budget for the transcript (12 of 20). The live interface is Socket.IO with an HTTP API beside it, and no headless command with machine-readable output was found (5 of 20). Checkpoints, clarification pauses that resume the same task, duplicate-call guards and background execution handles with wait and cancel (13 of 20). Three commands install and start it, a model provider has to be chosen at setup, and there are Node.js and Python SDKs for skills but no client SDK (7 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: Read with the self-hosted rule. No payment protocol (0). Free under MIT with nothing to buy, so 20, 20 and 20 on the last three lines. It runs with a local llama.cpp model and no sign-up. A hosted model is billed by its provider.

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, 60 out of 100, up to 3.5 more on the total

Made of editorial 52, provenance 68.

Why it scored 60: MIT (30). No privacy policy was found. getleon.ai says the owner controls their data and decides on any third party, the context documents describe per-profile isolation of memory, sessions and secrets, and neither the README nor the docs pages read mention the default telemetry (8 of 30). The README notice of 29 March 2026 says the project is on the 2.0 Developer Preview and points legacy users to `master`, and the docs site carries a banner that it is not being updated. No deprecation policy (8 of 20). Telemetry has a one-line opt-out in the sample config, `telemetry_enabled`, with no description of the fields sent or the retention, and an issue asking how to turn it off before install (#510, October 2023) is open with no reply (6 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: getleon.ai, registered 2018-02-16 (8 years) (11 of 15)
- Status page: 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.

- Kind is `harness`. Leon is a self-hosted agent program with no vendor API, which is closest to the harness kind the briefs define, and Schema, Ergonomics and Security use the harness reading of the Goose dossier.
- The edition graded is the 2.0 Developer Preview on `develop`, which the README names as the project's focus. The legacy 1.0.0-beta.8 on `master` was not graded.
- The lead says MIT and self-hosted, which is correct. Leon has no messaging-app channel and no MCP client on `develop`, so `assistant.messaging` and `agent.mcp-client` are not claimed.
- Judgement call. No deduction was taken for the default-on telemetry. A setting for it is in the sample config, we did not read every legacy docs page, and no dated incident was found.
- The telemetry module has a sender for utterance text. On `develop` the processor returns null or an empty object on the paths we read, so that sender is skipped. This was read in the source and not observed on a running instance.
- unchecked: whether GitHub private vulnerability reporting is enabled for the repository
- unchecked: the legacy docs pages other than the home and configuration pages, including whether any describes telemetry
- unchecked: roadmap.getleon.ai and changelog.getleon.ai, which the home page links over plain HTTP and which were not requested
- unchecked: what telemetry.getleon.ai stores and for how long. No request was sent to it
- unchecked: whether tests pass on `develop`. CI runs them only for `master`, and we did not run the software
- No legal entity, terms of service or privacy policy was found. The copyright holder is Louis Grenard
- What a month of use costs in model spend is not stated anywhere we read
- The `robots.txt` answers were github.com 200 (repository paths for tags, commits, tree and raw are closed to Anthropic agents, so tags came from git), getleon.ai 200 (only /api closed), blog.getleon.ai 200, docs.getleon.ai 404, api.github.com 404, registry.modelcontextprotocol.io 404, rdap.org 400, registry.npmjs.org returned a package record and no rules

## Weaknesses

- No tagged release of the core since 1.0.0-beta.8 on 1 May 2023. The 2.0 Developer Preview is installed by cloning the develop branch
- No approval step before a tool call. Shell commands pass a blocklist of patterns, and the proactive pulse is on by default and runs one task every 30 minutes
- The server listens on 0.0.0.0 and client authentication is off in the sample config
- Telemetry is on in the sample config and posts an instance ID, versions and hardware details to telemetry.getleon.ai. The README and the docs pages read do not describe it
- No `SECURITY.md`, security.txt, privacy policy or 2.0 documentation, and no messaging-app channel or MCP client on the develop branch

## 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.

- Clone the develop branch for 2.0. `npm install --global @leon-ai/cli` and docs.getleon.ai install and describe the legacy 1.0 betas
- Set `client_interface.auth.enabled: true` and a `LEON_PROFILE_TOKEN` before the machine joins a shared network. The server binds 0.0.0.0
- Set `telemetry_enabled: false` and `runtime.pulse_enabled: false` in the profile `config.yml` for an unattended or private run
- Use `availability.tools.allowed` to limit tools. Shell, file writes, computer use and browser control are otherwise available to the agent with no approval prompt
- Back up `LEON_CONNECTIONS_ENCRYPTION_KEY` from the profile `.env`. Without it the stored connections cannot be read

## 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.