Lemonade

# Fix list: Lemonade

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/lemonade, the October 2026 research run, assessed 8 October 2026. Grade B, 63.8 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 Lemonade: 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. Security & auth, 36 out of 100, up to 11.2 more on the total

Why it scored 36: Read with the tool checklist. Authentication is off until the operator sets `LEMONADE_API_KEY`, and with no key every route answers, including `/internal/*` shutdown and configuration. Two bearer keys set by environment variable separate regular from admin access, with no per-user keys and no rotation short of a restart, so 15 of 30, less 10 because the docs tell WebSocket clients to send the key as `?api_key=KEY` (5). The admin key fences off `/internal/*`, the default bind is localhost, cross-origin browser requests need an allow-list, MCP file writes are confined to a sandbox directory with traversal and symlink escapes rejected, and downloads need `allow_download`. The regular key still reaches model pull, delete and backend install, and there's no read-only mode (10 of 20). The tools return output from the owner's local models and no fetched web content. No prompt-injection guidance found in the docs (8 of 15). Optional OTLP traces for every inference request with session and client identifiers, Prometheus metrics at `/metrics` and a log stream WebSocket (11 of 15). GitHub reports no `SECURITY.md` and no advisories, lemonade-server.ai has no security.txt, NVD returned no entries for lemonade-sdk, and we found no disclosure address or bounty. Dependabot covers dev containers only (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.

## 2. Reliability, 75 out of 100, up to 5 more on the total

Why it scored 75: Read with the local-software lines, since Lemonade runs on the owner's hardware with no hosted service behind it. Official installers for each platform (Windows MSI and winget, macOS pkg, Ubuntu PPA and Snap, Debian deb, Fedora rpm, an Arch package and a container on ghcr.io), with supported systems and a backend-by-device table in the README (20). A public CI builds and tests every platform with C++ unit tests and Python server tests, and an MCP smoke test runs beside it. On 8 October 2026 the main build, test and release workflow's badge read failing on main, while the MCP smoke test and docs workflows read passing (15 of 25). 411 issues and 158 pull requests were open. New reports are labelled within a day or two, and open bugs from the past week include a server that stops responding after a client disconnects mid-request (#3769) and HTTP 500s from llama.cpp output parsing (#3770) (15 of 25). Versions are year.week.number since September, so the number carries no compatibility signal, but every release's notes have a Breaking Changes section, and most weekly releases list one or more (10 of 15). A stable channel separate from weekly candidates, and v11.9.0 before the scheme changed (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.

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

Why it scored 60: Read with the self-hosted rule. No x402, MPP or L402, and nothing is sold (0). Apache 2.0 with no account and no card, so 20, 20 and 20 on the last three lines. Cloud Offload uses the owner's own provider keys.

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.

## 4. Schema & documentation, 70 out of 100, up to 4.9 more on the total

Why it scored 70: We found no OpenAPI or Swagger file in the repository. The six MCP tools carry JSON Schema inputs from `tools/list`, though `messages` and `tools` are arrays of untyped objects and `stop` and `tool_choice` have empty schemas (12 of 25). No llms.txt at lemonade-server.ai (404). Every running server returns its own Markdown API reference at `GET /v1/docs` and through the `lemonade_docs` tool, so the reference matches the installed version and works offline (8 of 10). MCP tool descriptions say what to call first, which argument to prefer and what a wrong model name costs, and the API reference marks each OpenAI parameter as available or not (17 of 20). One enum (`response_format` on transcription), a minimum on `n` and required fields are declared, and the rest is typed loosely (8 of 15). Request examples in PowerShell and bash throughout, error tables on some endpoints and a JSON-RPC error table for MCP, but no single error reference (12 of 15). Notes with Headline and Breaking Changes sections on every release and versioned route prefixes, with no CHANGELOG file (13 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, 70 out of 100, up to 4.9 more on the total

Why it scored 70: Read with the MCP lines for `/mcp` and the API lines for the OpenAI-compatible routes. Six tools, with long descriptions (22 of 25). `max_tokens` caps output, `output_dir` returns file paths in place of inline base64, and `lemonade_list_models` has two include flags. We found no pagination on list routes (12 of 20). OpenAI-style error objects with a type and message, tool failures returned as `isError: true` with text a model can act on, and 501 on unsupported Ollama routes. No Retry-After or overload guidance found (15 of 20). No `readOnlyHint` or `destructiveHint` in the tool definitions and no idempotency keys. `allow_download` defaults to false, so a call can't start a download unless asked, and inference calls are safe to repeat (9 of 20). `model` is optional on the MCP tools, `lemonade launch claude` configures Claude Code in one command, and existing OpenAI, Anthropic and Ollama clients work unchanged, with no Lemonade client SDK (12 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. Transparency & trust, 64 out of 100, up to 3.2 more on the total

Made of editorial 70, provenance 57.

Why it scored 64: Apache 2.0, with backends downloaded separately under their own licences (30). We found no privacy policy for the software or the website. The README says the server is free and private, and the configuration guide documents the calls a default install makes (model and backend downloads, a start-up update check against Hugging Face, UDP broadcast for discovery) with switches for each. Nothing says what the project's own download hosts log (14 of 30). Release notes carry a Breaking Changes section and the release process is written down, but the one deprecation we found (`LEMONADE_ALLOWED_ORIGINS`) says only a future release, with no date (8 of 20). Telemetry is off by default, goes only to an OTLP endpoint the operator sets, and has documented switches to redact prompts, outputs and reasoning. A search of the source found no analytics library (18 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: lemonade-server.ai, registered 2025-05-12 (1 year) (3 of 15)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 76 out of 100, up to 2.1 more on the total

Why it scored 76: Stable release v2026.41.1 on 7 October 2026 (30). At least six stable releases in the 90 days to 8 October, from v11.7.0 on 19 August to v2026.41.1, with a candidate every week (20). New issues are labelled by area and engine within a day or two, 411 issues and 158 pull requests are open, and a Discord server is linked from the README. Comment threads weren't read (15 of 25). The official MCP registry returned no entry for Lemonade and there's no Lemonade client SDK, though OpenAI, Anthropic and Ollama SDKs work against it (5 of 15). A large CI matrix with self-hosted GPU and NPU runners, but the main workflow's badge read failing on 8 October and Dependabot is set up for dev containers only (6 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.

## 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.

- unchecked: who answers issues and how fast. Issue comment threads weren't read, and the GitHub API refused us for its rate limit
- unchecked: which job failed in the main build, test and release workflow on 8 October 2026. The runs list didn't render for our reader, so the failing state comes from the workflow badge alone
- unchecked: the first release date and the exact star count (the repository page showed 5.8k)
- unchecked: the licences of the separately downloaded backends, FastFlowLM and Ryzen AI among them
- unchecked: whether private vulnerability reporting is switched on for the repository
- unchecked: whether a `resources/list` or annotations field appears in a live `tools/list` response. We read the source and the generated docs and ran no server
- No terms of service or privacy policy was found for the software or the website, so `provenance.terms` and `provenance.privacy` are empty
- The legal entity is taken from the website footer (AMD) and `contrib/debian/copyright` (Advanced Micro Devices, Inc.). The `LICENSE` file has no copyright line

## Weaknesses

- No authentication by default. With no key set, every route answers, including `/internal/*` shutdown and configuration
- GitHub reports no `SECURITY.md`, and we found no security.txt, advisory or disclosure address
- The docs tell WebSocket clients to send the key as `?api_key=KEY` in the URL
- No OpenAPI file in the repository, and no `readOnlyHint` or `destructiveHint` on the MCP tools
- The main build and test workflow's badge read failing on main on 8 October 2026, with 411 issues open

## 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.

- Call `lemonade_list_models` (or `GET /v1/models`) before naming a model. A wrong name with `allow_download: true` can start a multi-gigabyte download
- Send `Authorization: Bearer <key>` when the operator has set `LEMONADE_API_KEY`. `/internal/*` needs the admin key when one is set
- Use `POST /v1/chat/completions` for streamed tokens, embeddings and speech. The MCP endpoint ignores `stream` and has no embeddings or text-to-speech tool
- Pass `output_dir` to `lemonade_generate_image` and `lemonade_omni` to get file paths in place of inline base64. Writes stay inside the MCP media sandbox
- Read `GET /v1/docs` on the running server for the reference that matches its version. Weekly releases change behaviour

## 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.