{ "data": { "similar": [ { "grade": "B", "json": "https://www.anchorterminal.com/tools/localai.json", "name": "LocalAI", "score": 68, "shared": [ "inference.local", "inference.open-weights" ], "slug": "localai" }, { "grade": "B", "json": "https://www.anchorterminal.com/tools/lemonade.json", "name": "Lemonade", "score": 63.8, "shared": [ "inference.local", "inference.open-weights" ], "slug": "lemonade" }, { "grade": "C", "json": "https://www.anchorterminal.com/tools/foundry-local.json", "name": "Foundry Local", "score": 60.5, "shared": [ "inference.local", "inference.open-weights" ], "slug": "foundry-local" }, { "grade": "C", "json": "https://www.anchorterminal.com/tools/koboldcpp.json", "name": "KoboldCpp", "score": 60.5, "shared": [ "inference.local", "inference.open-weights" ], "slug": "koboldcpp" }, { "grade": "C", "json": "https://www.anchorterminal.com/tools/llama-cpp.json", "name": "llama.cpp", "score": 60.2, "shared": [ "inference.local", "inference.open-weights" ], "slug": "llama-cpp" }, { "grade": "C", "json": "https://www.anchorterminal.com/tools/lm-studio.json", "name": "LM Studio", "score": 57.8, "shared": [ "inference.local", "inference.open-weights" ], "slug": "lm-studio" } ], "tool": { "slug": "mlx-lm", "name": "MLX LM", "vendor": "Apple Inc.", "vendorUrl": "https://opensource.apple.com/projects/mlx/", "kind": "http-api", "category": "local-ai", "summary": "Open-source Python package and command-line tools from Apple's MLX team for running, quantising and fine-tuning language models on Apple silicon. `mlx_lm.server` exposes a local HTTP API modelled on OpenAI's chat completions.", "url": "https://www.anchorterminal.com/tools/mlx-lm", "markdownUrl": "https://www.anchorterminal.com/tools/mlx-lm.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/mlx-lm.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/mlx-lm.json", "repo": "https://github.com/ml-explore/mlx-lm", "license": "MIT", "transports": [ "http" ], "packages": [ { "registry": "pypi", "name": "mlx-lm" } ], "auth": "none", "authNotes": "No credential, and no option to add one. `mlx_lm.server` binds 127.0.0.1:8080 by default, and `--allowed-origins` defaults to `*`, so any origin's requests are answered. Access control is left to the network or a proxy in front (https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md).", "pricing": "free", "pricingNotes": "Free under MIT, with no account, key or card. Nothing is sold. The owner pays for the hardware and electricity.", "priceSummary": "Free · OSS", "where": "local", "x402": { "level": "no", "evidence": "No x402, MPP or L402 in the docs or the source (checked 2026-10-08).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": 7300, "npmWeekly": null, "pypiWeekly": 139915, "asOf": "2026-10-08" }, "docsUrl": "https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md", "capabilities": [ "inference.local", "inference.open-weights" ], "tags": [ "open-source", "local", "self-hosted", "free", "no-card", "openai-compatible", "python", "pre-1.0", "no-auth", "no-telemetry" ], "lastRelease": "2026-10-01", "graded": true, "anchor": { "graded": true, "score": 52.2, "grade": "D", "agentReady": false, "rank": 657, "ranked": true, "rankOf": 842, "categoryRank": 12, "methodology": "0.4", "run": "2026-10-01", "scores": { "ergonomics": 54, "maintenance": 61, "payments": 60, "reliability": 66, "schema": 37, "security": 32, "transparency": 66 }, "pending": [ "performance", "tasks" ], "breakdown": [ { "key": "reliability", "name": "Reliability", "weight": 16, "effectiveWeight": 20, "score": 66, "points": 13.2, "reason": "Read with the local-software lines, since the package and its server run on the owner's machine. Installs from PyPI (`mlx-lm` 0.32.0) and conda-forge, with Python 3.11 or later stated and macOS and Linux classifiers (20). The Build and Test workflow runs a lint check and the unit tests (21 test files) on a macOS runner for every push to main, and the last eight runs on main passed (25). 151 open issues and 74 open pull requests. The newest bug reports have no reply or one comment, among them two on the server, #1946 (tool calls dropped from the response, 4 October) and #1909 (an uncaught exception that stops the generation thread, 21 September) (15 of 25). Version tags with GitHub release notes, no changelog file, and no breaking changes called out in the ten releases we read. The notes for v0.32.0 did not load (6 of 15). Version 0.32.0, pre-1.0, and the server docs say it is not recommended for production (0)." }, { "key": "performance", "name": "Performance", "weight": 10, "effectiveWeight": 0, "pending": true, "points": 0, "reason": "Pending. Latency is measured per call by our probes, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until the first probe window closes." }, { "key": "schema", "name": "Schema \u0026 documentation", "weight": 13, "effectiveWeight": 16.25, "score": 37, "points": 6.01, "reason": "Read for the HTTP server, the surface an agent calls. No OpenAPI file or other machine-readable contract. `SERVER.md` says the API is intended to be similar to OpenAI's chat API (5 of 25). No llms.txt. The docs are Markdown files in the repository, 175 lines for the server (5 of 10). Request and response fields each have a one-line purpose and the page opens with the production warning, but `tools`, `seed`, `max_completion_tokens`, `chat_template_kwargs`, the `/v1/completions` route and `/health` are in the code and not in the docs (9 of 20). Types and defaults are given in prose, and the server checks types and ranges in code. We found no `response_format` or JSON schema output in `server.py` (6 of 15). Two curl examples and no documented error responses (5 of 15). Version tags and GitHub release notes, with no changelog file and no version on the API itself beyond the `/v1` path (7 of 15)." }, { "key": "ergonomics", "name": "Agent ergonomics", "weight": 13, "effectiveWeight": 16.25, "score": 54, "points": 8.78, "reason": "Read for an API. `max_tokens` defaults to 512, `stop` and `logprobs` shape the output and `usage` reports token counts, but there is no field selection and no structured output mode (15 of 25). `GET /v1/models` lists the MLX models in the Hugging Face cache and can be narrowed to one repository by path. No paging and no token-counting route (10 of 20). Errors are `{\"error\": \"\u003ctext\u003e\"}` with 400 and a specific message for a bad field, 411 for a missing Content-Length, and 404 for any failure while loading a model or building the prompt. There are no codes and none of it is documented (8 of 20). Generation is stateless and safe to retry, `seed` is accepted, a prompt cache holds ten entries by default and `/health` answers 503 when the generation thread has stopped. No retry guidance (12 of 20). Only `messages` is required and the defaults are conservative. OpenAI clients can call the chat route, and the Python API (`load`, `generate`, `stream_generate`) ships in the same package. No client library in a second language in this repository (9 of 15)." }, { "key": "security", "name": "Security \u0026 auth", "weight": 14, "effectiveWeight": 17.5, "score": 32, "points": 5.6, "reason": "Read with the tool checklist. The server has no credential of any kind and no flag to add one. It binds 127.0.0.1 by default (5 of 30). `--allowed-origins` defaults to `*`, so a web page can call a server on localhost, and any request can name a `model` or `adapters` path for the server to download or load. `SERVER.md` says local paths must be relative to the start directory, and we found no such check in the code. `--trust-remote-code` is off by default (4 of 20). The server returns model output and parses tool calls without running them. No guidance on untrusted input (8 of 15). Logs at a chosen level, with request bodies at DEBUG, and no per-caller record (5 of 15). A security policy on the Security tab takes reports through GitHub private vulnerability reporting and says findings are published as advisories, the docs and a start-up warning say the server has only basic security checks, and no advisory is published. No bounty is named in the policy (10 of 20)." }, { "key": "payments", "name": "Payments \u0026 pricing", "weight": 10, "effectiveWeight": 12.5, "score": 60, "points": 7.5, "reason": "Read with the self-hosted rule. No x402, MPP or L402 in the docs or the source (0). Free under MIT with no account, key or card, and nothing to buy, so 20, 20 and 20 on the last three lines." }, { "key": "tasks", "name": "Task success", "weight": 10, "effectiveWeight": 0, "pending": true, "points": 0, "reason": "Pending. Task success needs the category task suites run through each tool, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until then. A data provider's data-quality score is published on its listing now and becomes half of this category when it's scored." }, { "key": "maintenance", "name": "Maintenance \u0026 community", "weight": 7, "effectiveWeight": 8.75, "score": 61, "points": 5.34, "reason": "PyPI release 0.32.0 on 1 October 2026 (30). That is the only release in 90 days, the one before being 0.31.3 on 22 April 2026 (0). 127 commits from 82 authors on main since 10 July, with commits on most days of the last week. 151 open issues and 74 open pull requests, and recent bug reports mostly have no reply or one comment. Reply times were not measured (15 of 25). The package is the Python library, current with the release. No other official client (8 of 15). CI on every push, a pre-commit lint job, PyPI publishing by trusted publishing, and dependency floors on recent versions (`mlx\u003e=0.32.2`, `transformers\u003e=5.7.0`). No Dependabot file (8 of 10)." }, { "key": "transparency", "name": "Transparency \u0026 trust", "weight": 7, "effectiveWeight": 8.75, "score": 66, "points": 5.78, "note": "editorial 65, provenance 67", "reason": "The editorial half. MIT, copyright Apple Inc., all of it public (30). No privacy statement covers the software and nothing states what it sends. In the source the outbound calls are model downloads from Hugging Face (or ModelScope when `MLXLM_USE_MODELSCOPE` is set), uploads only on command, and a calibration text fetched from a GitHub gist by the quantisation tools (12 of 30). Runtime messages say `python -m mlx_lm.\u003ccommand\u003e` is deprecated, with no date and no written policy (5 of 20). No telemetry, analytics or update check found in the source, so nothing to opt out of, though the README does not say so (18 of 20)." } ], "assessment": { "date": "2026-10-08", "basis": "public evidence", "confidence": "medium", "notes": { "ergonomics": "Read for an API. `max_tokens` defaults to 512, `stop` and `logprobs` shape the output and `usage` reports token counts, but there is no field selection and no structured output mode (15 of 25). `GET /v1/models` lists the MLX models in the Hugging Face cache and can be narrowed to one repository by path. No paging and no token-counting route (10 of 20). Errors are `{\"error\": \"\u003ctext\u003e\"}` with 400 and a specific message for a bad field, 411 for a missing Content-Length, and 404 for any failure while loading a model or building the prompt. There are no codes and none of it is documented (8 of 20). Generation is stateless and safe to retry, `seed` is accepted, a prompt cache holds ten entries by default and `/health` answers 503 when the generation thread has stopped. No retry guidance (12 of 20). Only `messages` is required and the defaults are conservative. OpenAI clients can call the chat route, and the Python API (`load`, `generate`, `stream_generate`) ships in the same package. No client library in a second language in this repository (9 of 15).", "maintenance": "PyPI release 0.32.0 on 1 October 2026 (30). That is the only release in 90 days, the one before being 0.31.3 on 22 April 2026 (0). 127 commits from 82 authors on main since 10 July, with commits on most days of the last week. 151 open issues and 74 open pull requests, and recent bug reports mostly have no reply or one comment. Reply times were not measured (15 of 25). The package is the Python library, current with the release. No other official client (8 of 15). CI on every push, a pre-commit lint job, PyPI publishing by trusted publishing, and dependency floors on recent versions (`mlx\u003e=0.32.2`, `transformers\u003e=5.7.0`). No Dependabot file (8 of 10).", "payments": "Read with the self-hosted rule. No x402, MPP or L402 in the docs or the source (0). Free under MIT with no account, key or card, and nothing to buy, so 20, 20 and 20 on the last three lines.", "reliability": "Read with the local-software lines, since the package and its server run on the owner's machine. Installs from PyPI (`mlx-lm` 0.32.0) and conda-forge, with Python 3.11 or later stated and macOS and Linux classifiers (20). The Build and Test workflow runs a lint check and the unit tests (21 test files) on a macOS runner for every push to main, and the last eight runs on main passed (25). 151 open issues and 74 open pull requests. The newest bug reports have no reply or one comment, among them two on the server, #1946 (tool calls dropped from the response, 4 October) and #1909 (an uncaught exception that stops the generation thread, 21 September) (15 of 25). Version tags with GitHub release notes, no changelog file, and no breaking changes called out in the ten releases we read. The notes for v0.32.0 did not load (6 of 15). Version 0.32.0, pre-1.0, and the server docs say it is not recommended for production (0).", "schema": "Read for the HTTP server, the surface an agent calls. No OpenAPI file or other machine-readable contract. `SERVER.md` says the API is intended to be similar to OpenAI's chat API (5 of 25). No llms.txt. The docs are Markdown files in the repository, 175 lines for the server (5 of 10). Request and response fields each have a one-line purpose and the page opens with the production warning, but `tools`, `seed`, `max_completion_tokens`, `chat_template_kwargs`, the `/v1/completions` route and `/health` are in the code and not in the docs (9 of 20). Types and defaults are given in prose, and the server checks types and ranges in code. We found no `response_format` or JSON schema output in `server.py` (6 of 15). Two curl examples and no documented error responses (5 of 15). Version tags and GitHub release notes, with no changelog file and no version on the API itself beyond the `/v1` path (7 of 15).", "security": "Read with the tool checklist. The server has no credential of any kind and no flag to add one. It binds 127.0.0.1 by default (5 of 30). `--allowed-origins` defaults to `*`, so a web page can call a server on localhost, and any request can name a `model` or `adapters` path for the server to download or load. `SERVER.md` says local paths must be relative to the start directory, and we found no such check in the code. `--trust-remote-code` is off by default (4 of 20). The server returns model output and parses tool calls without running them. No guidance on untrusted input (8 of 15). Logs at a chosen level, with request bodies at DEBUG, and no per-caller record (5 of 15). A security policy on the Security tab takes reports through GitHub private vulnerability reporting and says findings are published as advisories, the docs and a start-up warning say the server has only basic security checks, and no advisory is published. No bounty is named in the policy (10 of 20).", "transparency": "The editorial half. MIT, copyright Apple Inc., all of it public (30). No privacy statement covers the software and nothing states what it sends. In the source the outbound calls are model downloads from Hugging Face (or ModelScope when `MLXLM_USE_MODELSCOPE` is set), uploads only on command, and a calibration text fetched from a GitHub gist by the quantisation tools (12 of 30). Runtime messages say `python -m mlx_lm.\u003ccommand\u003e` is deprecated, with no date and no written policy (5 of 20). No telemetry, analytics or update check found in the source, so nothing to opt out of, though the README does not say so (18 of 20)." }, "sources": [ { "what": "repository README, licence and header counts", "url": "https://github.com/ml-explore/mlx-lm", "seen": "2026-10-08" }, { "what": "server documentation", "url": "https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md", "seen": "2026-10-08" }, { "what": "server source (routes, flags, CORS, errors)", "url": "https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/server.py", "seen": "2026-10-08" }, { "what": "model download and path handling", "url": "https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/utils.py", "seen": "2026-10-08" }, { "what": "packaging, supported Python and dependencies", "url": "https://github.com/ml-explore/mlx-lm/blob/main/pyproject.toml", "seen": "2026-10-08" }, { "what": "CI workflow", "url": "https://github.com/ml-explore/mlx-lm/blob/main/.github/workflows/pull_request.yml", "seen": "2026-10-08" }, { "what": "CI runs on main", "url": "https://github.com/ml-explore/mlx-lm/actions/workflows/pull_request.yml?query=branch%3Amain", "seen": "2026-10-08" }, { "what": "releases", "url": "https://github.com/ml-explore/mlx-lm/releases", "seen": "2026-10-08" }, { "what": "PyPI release history", "url": "https://pypi.org/pypi/mlx-lm/json", "seen": "2026-10-08" }, { "what": "PyPI download counts", "url": "https://pypistats.org/api/packages/mlx-lm/recent", "seen": "2026-10-08" }, { "what": "open issues", "url": "https://github.com/ml-explore/mlx-lm/issues", "seen": "2026-10-08" }, { "what": "issue asking for a model allow-list", "url": "https://github.com/ml-explore/mlx-lm/issues/1892", "seen": "2026-10-08" }, { "what": "security policy", "url": "https://github.com/ml-explore/mlx-lm/security/policy", "seen": "2026-10-08" }, { "what": "security advisories", "url": "https://github.com/ml-explore/mlx-lm/security", "seen": "2026-10-08" }, { "what": "instructions addressed to AI coding assistants", "url": "https://github.com/ml-explore/mlx-lm/blob/main/AGENTS.md", "seen": "2026-10-08" }, { "what": "contributing guide and AI usage policy", "url": "https://github.com/ml-explore/mlx-lm/blob/main/CONTRIBUTING.md", "seen": "2026-10-08" }, { "what": "calibration data download", "url": "https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/quant/utils.py", "seen": "2026-10-08" }, { "what": "Apple open source MLX page", "url": "https://opensource.apple.com/projects/mlx/", "seen": "2026-10-08" }, { "what": "Apple security.txt", "url": "https://www.apple.com/.well-known/security.txt", "seen": "2026-10-08" } ], "openQuestions": [ "unchecked: the GitHub release notes for v0.32.0, which did not load. The tag (30 September 2026) and the PyPI upload (1 October) were read", "unchecked: reply times on issues and pull requests. Only comment counts on the newest open issues were read", "unchecked: whether Apple's security bounty covers mlx-lm. The repository's policy names GitHub private reporting only", "unchecked: the conda-forge package version. The page answered 200 and was not read", "No terms or privacy document governs the software, so `provenance.terms` and `provenance.privacy` are empty. The pages linked from opensource.apple.com are Apple's website terms and general privacy policy", "Whether the server enforces the documented rule that a local model path is relative to the start directory. We found no check in `server.py` or `utils.py` and did not run the server" ] }, "negative": 0, "verdict": "MIT, with no telemetry found in the source, and the tests passed on the last eight pushes to main. `mlx_lm.server` has no API key option, answers any origin by default and loads whichever model a request names, and its own docs say it is not recommended for production.", "bestFor": "An owner with an Apple silicon Mac who wants MLX-format models, local fine-tuning and quantisation from Python or the command line, with a simple local chat completions server.", "strengths": [ "MIT, with no telemetry, analytics or update check found in the source", "Installs from PyPI (`mlx-lm` 0.32.0, Python 3.11 or later) and conda-forge, with releases published to PyPI by trusted publishing from a GitHub workflow", "The Build and Test workflow passed on the last eight pushes to main, with 21 test files run on a macOS runner", "`mlx_lm.server` binds 127.0.0.1:8080 by default, caps output at 512 tokens unless told otherwise and validates field types and ranges with a 400", "127 commits from 82 authors on main in the 90 days to 8 October 2026" ], "weaknesses": [ "`mlx_lm.server` has no API key or other credential option, and `--allowed-origins` defaults to `*`", "A request's `model` and `adapters` fields make the server download or load any Hugging Face repository or local path, with no allow-list (open issue #1892)", "The docs and a start-up warning say the server is not recommended for production because it has only basic security checks", "No OpenAPI file or llms.txt, and `SERVER.md` leaves out `tools`, `seed`, `/health` and the error responses", "One PyPI release in 90 days (0.32.0 on 1 October 2026, the first since 0.31.3 on 22 April), and the version is still 0.x" ], "agentNotes": [ "Keep `mlx_lm.server` on 127.0.0.1 and pass `--allowed-origins` with the origins you trust. There is no API key, and the default answers every origin", "Treat any caller as able to load any model. The `model` and `adapters` request fields accept any Hugging Face repository or local path", "Send `max_tokens` or `max_completion_tokens` when you need more than 512 tokens, the server default", "Read errors as `{\"error\": \"\u003ctext\u003e\"}` with 400 for a bad field and 404 for a model that failed to load. They are not OpenAI error objects", "Poll `GET /health` before the first request. It answers 503 with `unavailable` when the generation thread has stopped" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "D", "methodology": "0.4", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 52.2 } ], "editorialScores": { "ergonomics": 54, "maintenance": 61, "payments": 60, "reliability": 66, "schema": 37, "security": 32, "transparency": 65 }, "provenanceScore": 67 }, "connect": { "install": "pip install mlx-lm\nmlx_lm.server --model mlx-community/Mistral-7B-Instruct-v0.3-4bit # listens on 127.0.0.1:8080", "http": "curl localhost:8080/v1/chat/completions \\\n -H \"Content-Type: application/json\" \\\n -d '{\n \"messages\": [{\"role\": \"user\", \"content\": \"Say this is a test!\"}],\n \"temperature\": 0.7\n }'" }, "letme": { "capability": "https://letme.dev/inference.local", "tool": "https://letme.dev/mlx-lm" }, "notable": [ "The server docs open with a note that the MLX LM server is not recommended for production as it only implements basic security checks, and the same text is a warning at start-up (https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md)", "`mlx_lm.server` has no credential option. Its flags cover host, port and `--allowed-origins`, which defaults to `*` (https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/server.py)", "A request's `model` field is passed to the loader, which uses a local path if it exists and otherwise downloads the repository from Hugging Face. `SERVER.md` says a local path must be relative to the start directory, and we found no such check in `server.py` or `utils.py`. Issue #1892 of 15 September 2026 asks for an allow-list and has no reply (https://github.com/ml-explore/mlx-lm/issues/1892)", "The repository's `AGENTS.md` and `CLAUDE.md` are instructions addressed to AI coding assistants about contributions (no automated pull requests, replies or commit messages). They concern contributors, not users of the package, and we did not act on them (https://github.com/ml-explore/mlx-lm/blob/main/AGENTS.md)", "Tag v0.32.0 is dated 30 September 2026 and PyPI has 0.32.0 from 1 October, the first release since 0.31.3 on 22 April 2026 (https://pypi.org/project/mlx-lm/)", "The Security tab shows a policy that takes reports through GitHub private vulnerability reporting and publishes findings as advisories. No advisory is published (https://github.com/ml-explore/mlx-lm/security)", "The quantisation tools download a calibration text from a personal GitHub gist on first use and keep it in `~/.cache/mlx-lm` (https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/quant/utils.py)" ], "area": "models", "details": [ { "label": "Interfaces", "value": "`mlx_lm.server` HTTP API, the Python API (`load`, `generate`, `stream_generate`, `convert`), and command-line tools including `mlx_lm.generate`, `mlx_lm.chat`, `mlx_lm.lora`, `mlx_lm.convert`, `mlx_lm.fuse`, `mlx_lm.evaluate`, `mlx_lm.cache_prompt` and `mlx_lm.manage`" }, { "label": "Routes", "value": "POST `/v1/chat/completions` (also `/chat/completions`) and `/v1/completions`, GET `/v1/models` and `/health`" }, { "label": "Credentials", "value": "None, and no flag to add one" }, { "label": "Network defaults", "value": "Binds 127.0.0.1:8080. `--allowed-origins` defaults to `*`. Models download from Hugging Face on first use, or from ModelScope when `MLXLM_USE_MODELSCOPE` is true" }, { "label": "Request defaults", "value": "`max_tokens` 512, `temperature` 0.0, `top_p` 1.0. Only `messages` is required" }, { "label": "Models", "value": "MLX-format models from Hugging Face, with 4-bit and other quantised builds in the mlx-community organisation. The default for `mlx_lm.generate` and `mlx_lm.chat` is `mlx-community/Llama-3.2-3B-Instruct-4bit`" }, { "label": "Tool calling", "value": "The chat route accepts `tools` and parses tool calls with one of 13 model-specific parsers. It does not run them, and `SERVER.md` does not document the field" }, { "label": "Fine-tuning", "value": "Low-rank (LoRA) and full fine-tuning with `mlx_lm.lora`, including on quantised models, and adapters fused with `mlx_lm.fuse`" }, { "label": "Platforms", "value": "Python 3.11 to 3.13. macOS on Apple silicon, with Linux classifiers and `cuda12`, `cuda13` and `cpu` extras. Wired memory for large models needs macOS 15 or later" }, { "label": "Install", "value": "`pip install mlx-lm` or `conda install -c conda-forge mlx-lm`" }, { "label": "Releases", "value": "97 versions on PyPI since 0.0.1 on 12 January 2024. 0.32.0 on 1 October 2026, the one before 0.31.3 on 22 April 2026" }, { "label": "Activity", "value": "127 commits from 82 authors on main in the 90 days to 8 October 2026. 151 open issues and 74 open pull requests" }, { "label": "Security record", "value": "No published GitHub advisory. Reports go through GitHub private vulnerability reporting" } ], "provenance": { "legalEntity": "Apple Inc.", "domain": "apple.com", "domainRegistered": "", "endpointOnVendorDomain": null, "terms": "", "privacy": "", "statusPage": "", "changelog": "https://github.com/ml-explore/mlx-lm/releases", "securityTxt": "valid", "checked": "2026-10-08", "notes": [ "The `LICENSE` file reads Copyright 2023 Apple Inc., and the package author on PyPI is MLX Contributors at a group.apple.com address. The repository sits in GitHub's ml-explore organisation and has no website of its own.", "opensource.apple.com/projects/mlx describes the MLX framework and does not name MLX LM. Its footer links Apple's website terms and general privacy policy, which do not govern this software, so terms and privacy are left empty.", "www.apple.com/.well-known/security.txt is valid until 6 October 2027 and is Apple's corporate file. The repository's own policy takes reports through GitHub private vulnerability reporting.", "There is no shared hosted endpoint. The server runs on the owner's machine." ], "score": 67, "checks": [ { "check": "Legal entity named", "value": "Apple Inc.", "points": 20, "max": 20, "state": "ok" }, { "check": "Domain age", "value": "apple.com, no registry record we could read", "points": 0, "max": 15, "state": "no" }, { "check": "Endpoint on the vendor's domain", "value": "no hosted endpoint", "points": 0, "max": 0, "state": "na" }, { "check": "Terms of service", "value": "nothing hosted, so the MIT licence stands in", "points": 10, "max": 10, "state": "ok" }, { "check": "Privacy policy", "value": "nothing hosted, not scored", "points": 0, "max": 0, "state": "na" }, { "check": "Status page", "value": "not found", "points": 0, "max": 10, "state": "no" }, { "check": "Changelog", "value": "published", "points": 10, "max": 10, "state": "ok" }, { "check": "security.txt", "value": "valid", "points": 10, "max": 10, "state": "ok" } ] }, "pageJsonUrl": "https://www.anchorterminal.com/tools/mlx-lm.json" }, "verify": { "accepts": "a page on apple.com or one of its subdomains, or the README of github.com/ml-explore/mlx-lm", "badgeUrl": "https://www.anchorterminal.com/badges/mlx-lm.svg", "body": { "slug": "mlx-lm", "url": "the page with the badge or the link" }, "docs": "https://www.anchorterminal.com/builders/#verify", "effect": "none, it never changes a grade, rank or review", "endpoint": "https://www.anchorterminal.com/api/v1/verify", "listingUrl": "https://www.anchorterminal.com/tools/mlx-lm", "mcpTool": "verify_listing", "recheck": "weekly; two failed checks in a row and it lapses, a later pass restores it", "snippets": { "html": "\u003ca href=\"https://www.anchorterminal.com/tools/mlx-lm\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/mlx-lm.svg\" alt=\"MLX LM on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e", "markdown": "[![MLX LM on Anchor Terminal](https://www.anchorterminal.com/badges/mlx-lm.svg)](https://www.anchorterminal.com/tools/mlx-lm)", "link": "\u003ca href=\"https://www.anchorterminal.com/tools/mlx-lm\"\u003eMLX LM on Anchor Terminal\u003c/a\u003e" } } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/tools/mlx-lm", "json": "https://www.anchorterminal.com/tools/mlx-lm.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/tools/mlx-lm.md", "slim": "https://www.anchorterminal.com/tools/mlx-lm.min.md" }, "markdown": "## Overview\n\n**Grade D · 52.2/100 · rank #657 of 842 · #12 in Local AI · not agent-ready · confidence medium**\n\n\n## Assessment\n\nMIT, with no telemetry found in the source, and the tests passed on the last eight pushes to main. `mlx_lm.server` has no API key option, answers any origin by default and loads whichever model a request names, and its own docs say it is not recommended for production.\n\n## Facts\n\n| Field | Value |\n| --- | --- |\n| Vendor | Apple Inc. (https://opensource.apple.com/projects/mlx/) |\n| Kind | HTTP API |\n| Category | Local AI (https://www.anchorterminal.com/categories/local-ai) |\n| Transport | HTTP |\n| Auth | None · No credential, and no option to add one. `mlx_lm.server` binds 127.0.0.1:8080 by default, and `--allowed-origins` defaults to `*`, so any origin's requests are answered. Access control is left to the network or a proxy in front (https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md). |\n| Pricing | Free (Free · OSS) · Free under MIT, with no account, key or card. Nothing is sold. The owner pays for the hardware and electricity. |\n| x402 | No · No x402, MPP or L402 in the docs or the source (checked 2026-10-08). |\n| Licence | MIT |\n| Packages | pypi: `mlx-lm` |\n| Source | https://github.com/ml-explore/mlx-lm |\n| Docs | https://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md |\n| llms.txt | not found |\n| Last release | 2026-10-01 |\n| GitHub stars | 7,300 (as of 2026-10-08) |\n| PyPI downloads / week | 139,915 |\n| Interfaces | `mlx_lm.server` HTTP API, the Python API (`load`, `generate`, `stream_generate`, `convert`), and command-line tools including `mlx_lm.generate`, `mlx_lm.chat`, `mlx_lm.lora`, `mlx_lm.convert`, `mlx_lm.fuse`, `mlx_lm.evaluate`, `mlx_lm.cache_prompt` and `mlx_lm.manage` |\n| Routes | POST `/v1/chat/completions` (also `/chat/completions`) and `/v1/completions`, GET `/v1/models` and `/health` |\n| Credentials | None, and no flag to add one |\n| Network defaults | Binds 127.0.0.1:8080. `--allowed-origins` defaults to `*`. Models download from Hugging Face on first use, or from ModelScope when `MLXLM_USE_MODELSCOPE` is true |\n| Request defaults | `max_tokens` 512, `temperature` 0.0, `top_p` 1.0. Only `messages` is required |\n| Models | MLX-format models from Hugging Face, with 4-bit and other quantised builds in the mlx-community organisation. The default for `mlx_lm.generate` and `mlx_lm.chat` is `mlx-community/Llama-3.2-3B-Instruct-4bit` |\n| Tool calling | The chat route accepts `tools` and parses tool calls with one of 13 model-specific parsers. It does not run them, and `SERVER.md` does not document the field |\n| Fine-tuning | Low-rank (LoRA) and full fine-tuning with `mlx_lm.lora`, including on quantised models, and adapters fused with `mlx_lm.fuse` |\n| Platforms | Python 3.11 to 3.13. macOS on Apple silicon, with Linux classifiers and `cuda12`, `cuda13` and `cpu` extras. Wired memory for large models needs macOS 15 or later |\n| Install | `pip install mlx-lm` or `conda install -c conda-forge mlx-lm` |\n| Releases | 97 versions on PyPI since 0.0.1 on 12 January 2024. 0.32.0 on 1 October 2026, the one before 0.31.3 on 22 April 2026 |\n| Activity | 127 commits from 82 authors on main in the 90 days to 8 October 2026. 151 open issues and 74 open pull requests |\n| Security record | No published GitHub advisory. Reports go through GitHub private vulnerability reporting |\n| Capabilities | inference.local, inference.open-weights |\n| Tags | open-source, local, self-hosted, free, no-card, openai-compatible, python, pre-1.0, no-auth, no-telemetry |\n| JSON | https://www.anchorterminal.com/api/v1/tools/mlx-lm.json |\n\n## Score breakdown (methodology v0.4, October 2026 research run)\n\nAssessed 2026-10-08 from public evidence against the published checklist (https://www.anchorterminal.com/benchmark/#checklist). Confidence: medium. Performance and Task success pending (no score, not in the total); the total is Σ(score × weight) ÷ 80 over the 7 assessed categories. \"This run\" is each category's share of the 100 points.\n\n| Category | Weight | This run | Score (0–100) | Points |\n| --- | --- | --- | --- | --- |\n| Reliability | 16% | 20 | 66 | 13.2 |\n| Performance | 10% | pending | pending | n/a |\n| Schema \u0026 documentation | 13% | 16.2 | 37 | 6.0 |\n| Agent ergonomics | 13% | 16.2 | 54 | 8.8 |\n| Security \u0026 auth | 14% | 17.5 | 32 | 5.6 |\n| Payments \u0026 pricing | 10% | 12.5 | 60 | 7.5 |\n| Task success | 10% | pending | pending | n/a |\n| Maintenance \u0026 community | 7% | 8.8 | 61 | 5.3 |\n| Transparency \u0026 trust (editorial 65, provenance 67) | 7% | 8.8 | 66 | 5.8 |\n| Negative events | up to −15 | up to −15 | none recorded | 0 |\n| **Total** | | | | **52.2 → D** |\n\n### Why each score\n\n- Reliability 66: Read with the local-software lines, since the package and its server run on the owner's machine. Installs from PyPI (`mlx-lm` 0.32.0) and conda-forge, with Python 3.11 or later stated and macOS and Linux classifiers (20). The Build and Test workflow runs a lint check and the unit tests (21 test files) on a macOS runner for every push to main, and the last eight runs on main passed (25). 151 open issues and 74 open pull requests. The newest bug reports have no reply or one comment, among them two on the server, #1946 (tool calls dropped from the response, 4 October) and #1909 (an uncaught exception that stops the generation thread, 21 September) (15 of 25). Version tags with GitHub release notes, no changelog file, and no breaking changes called out in the ten releases we read. The notes for v0.32.0 did not load (6 of 15). Version 0.32.0, pre-1.0, and the server docs say it is not recommended for production (0).\n- Performance: Pending. Latency is measured per call by our probes, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until the first probe window closes.\n- Schema \u0026 documentation 37: Read for the HTTP server, the surface an agent calls. No OpenAPI file or other machine-readable contract. `SERVER.md` says the API is intended to be similar to OpenAI's chat API (5 of 25). No llms.txt. The docs are Markdown files in the repository, 175 lines for the server (5 of 10). Request and response fields each have a one-line purpose and the page opens with the production warning, but `tools`, `seed`, `max_completion_tokens`, `chat_template_kwargs`, the `/v1/completions` route and `/health` are in the code and not in the docs (9 of 20). Types and defaults are given in prose, and the server checks types and ranges in code. We found no `response_format` or JSON schema output in `server.py` (6 of 15). Two curl examples and no documented error responses (5 of 15). Version tags and GitHub release notes, with no changelog file and no version on the API itself beyond the `/v1` path (7 of 15).\n- Agent ergonomics 54: Read for an API. `max_tokens` defaults to 512, `stop` and `logprobs` shape the output and `usage` reports token counts, but there is no field selection and no structured output mode (15 of 25). `GET /v1/models` lists the MLX models in the Hugging Face cache and can be narrowed to one repository by path. No paging and no token-counting route (10 of 20). Errors are `{\"error\": \"\u003ctext\u003e\"}` with 400 and a specific message for a bad field, 411 for a missing Content-Length, and 404 for any failure while loading a model or building the prompt. There are no codes and none of it is documented (8 of 20). Generation is stateless and safe to retry, `seed` is accepted, a prompt cache holds ten entries by default and `/health` answers 503 when the generation thread has stopped. No retry guidance (12 of 20). Only `messages` is required and the defaults are conservative. OpenAI clients can call the chat route, and the Python API (`load`, `generate`, `stream_generate`) ships in the same package. No client library in a second language in this repository (9 of 15).\n- Security \u0026 auth 32: Read with the tool checklist. The server has no credential of any kind and no flag to add one. It binds 127.0.0.1 by default (5 of 30). `--allowed-origins` defaults to `*`, so a web page can call a server on localhost, and any request can name a `model` or `adapters` path for the server to download or load. `SERVER.md` says local paths must be relative to the start directory, and we found no such check in the code. `--trust-remote-code` is off by default (4 of 20). The server returns model output and parses tool calls without running them. No guidance on untrusted input (8 of 15). Logs at a chosen level, with request bodies at DEBUG, and no per-caller record (5 of 15). A security policy on the Security tab takes reports through GitHub private vulnerability reporting and says findings are published as advisories, the docs and a start-up warning say the server has only basic security checks, and no advisory is published. No bounty is named in the policy (10 of 20).\n- Payments \u0026 pricing 60: Read with the self-hosted rule. No x402, MPP or L402 in the docs or the source (0). Free under MIT with no account, key or card, and nothing to buy, so 20, 20 and 20 on the last three lines.\n- Task success: Pending. Task success needs the category task suites run through each tool, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until then. A data provider's data-quality score is published on its listing now and becomes half of this category when it's scored.\n- Maintenance \u0026 community 61: PyPI release 0.32.0 on 1 October 2026 (30). That is the only release in 90 days, the one before being 0.31.3 on 22 April 2026 (0). 127 commits from 82 authors on main since 10 July, with commits on most days of the last week. 151 open issues and 74 open pull requests, and recent bug reports mostly have no reply or one comment. Reply times were not measured (15 of 25). The package is the Python library, current with the release. No other official client (8 of 15). CI on every push, a pre-commit lint job, PyPI publishing by trusted publishing, and dependency floors on recent versions (`mlx\u003e=0.32.2`, `transformers\u003e=5.7.0`). No Dependabot file (8 of 10).\n- Transparency \u0026 trust 66: The editorial half. MIT, copyright Apple Inc., all of it public (30). No privacy statement covers the software and nothing states what it sends. In the source the outbound calls are model downloads from Hugging Face (or ModelScope when `MLXLM_USE_MODELSCOPE` is set), uploads only on command, and a calibration text fetched from a GitHub gist by the quantisation tools (12 of 30). Runtime messages say `python -m mlx_lm.\u003ccommand\u003e` is deprecated, with no date and no written policy (5 of 20). No telemetry, analytics or update check found in the source, so nothing to opt out of, though the README does not say so (18 of 20).\n\nFix list for a coding agent, everything this grade says the listing lacks, the biggest gain first (15 items): https://www.anchorterminal.com/fixes/mlx-lm.md (JSON https://www.anchorterminal.com/fixes/mlx-lm.json)\n\n### What we couldn't check\n\n- unchecked: the GitHub release notes for v0.32.0, which did not load. The tag (30 September 2026) and the PyPI upload (1 October) were read\n- unchecked: reply times on issues and pull requests. Only comment counts on the newest open issues were read\n- unchecked: whether Apple's security bounty covers mlx-lm. The repository's policy names GitHub private reporting only\n- unchecked: the conda-forge package version. The page answered 200 and was not read\n- No terms or privacy document governs the software, so `provenance.terms` and `provenance.privacy` are empty. The pages linked from opensource.apple.com are Apple's website terms and general privacy policy\n- Whether the server enforces the documented rule that a local model path is relative to the start directory. We found no check in `server.py` or `utils.py` and did not run the server\n\n### Sources\n\n- repository README, licence and header counts: \u003chttps://github.com/ml-explore/mlx-lm\u003e (seen 2026-10-08)\n- server documentation: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md\u003e (seen 2026-10-08)\n- server source (routes, flags, CORS, errors): \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/server.py\u003e (seen 2026-10-08)\n- model download and path handling: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/utils.py\u003e (seen 2026-10-08)\n- packaging, supported Python and dependencies: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/pyproject.toml\u003e (seen 2026-10-08)\n- CI workflow: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/.github/workflows/pull_request.yml\u003e (seen 2026-10-08)\n- CI runs on main: \u003chttps://github.com/ml-explore/mlx-lm/actions/workflows/pull_request.yml?query=branch%3Amain\u003e (seen 2026-10-08)\n- releases: \u003chttps://github.com/ml-explore/mlx-lm/releases\u003e (seen 2026-10-08)\n- PyPI release history: \u003chttps://pypi.org/pypi/mlx-lm/json\u003e (seen 2026-10-08)\n- PyPI download counts: \u003chttps://pypistats.org/api/packages/mlx-lm/recent\u003e (seen 2026-10-08)\n- open issues: \u003chttps://github.com/ml-explore/mlx-lm/issues\u003e (seen 2026-10-08)\n- issue asking for a model allow-list: \u003chttps://github.com/ml-explore/mlx-lm/issues/1892\u003e (seen 2026-10-08)\n- security policy: \u003chttps://github.com/ml-explore/mlx-lm/security/policy\u003e (seen 2026-10-08)\n- security advisories: \u003chttps://github.com/ml-explore/mlx-lm/security\u003e (seen 2026-10-08)\n- instructions addressed to AI coding assistants: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/AGENTS.md\u003e (seen 2026-10-08)\n- contributing guide and AI usage policy: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/CONTRIBUTING.md\u003e (seen 2026-10-08)\n- calibration data download: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/quant/utils.py\u003e (seen 2026-10-08)\n- Apple open source MLX page: \u003chttps://opensource.apple.com/projects/mlx/\u003e (seen 2026-10-08)\n- Apple security.txt: \u003chttps://www.apple.com/.well-known/security.txt\u003e (seen 2026-10-08)\n\n## Who's behind it (provenance 67/100, checked 2026-10-08)\n\n| Check | Finding | Points |\n| --- | --- | --- |\n| Legal entity named | Apple Inc. | 20/20 |\n| Domain age | apple.com, no registry record we could read | 0/15 |\n| Endpoint on the vendor's domain | no hosted endpoint | n/a |\n| Terms of service | nothing hosted, so the MIT licence stands in | 10/10 |\n| Privacy policy | nothing hosted, not scored | n/a |\n| Status page | not found | 0/10 |\n| Changelog | published | 10/10 |\n| security.txt | valid | 10/10 |\n\nThe `LICENSE` file reads Copyright 2023 Apple Inc., and the package author on PyPI is MLX Contributors at a group.apple.com address. The repository sits in GitHub's ml-explore organisation and has no website of its own.\n\nopensource.apple.com/projects/mlx describes the MLX framework and does not name MLX LM. Its footer links Apple's website terms and general privacy policy, which do not govern this software, so terms and privacy are left empty.\n\nwww.apple.com/.well-known/security.txt is valid until 6 October 2027 and is Apple's corporate file. The repository's own policy takes reports through GitHub private vulnerability reporting.\n\nThere is no shared hosted endpoint. The server runs on the owner's machine.\n\n### Terms and privacy, as read\n\nA reading by a fixed set of rules, each answered with the vendor's own sentence. Not legal advice.\n\n**Terms of service**. Nothing is hosted by the vendor, so there are no terms of service to read. The MIT licence stands in and the check scores in full.\n\n\n**Privacy policy**. Nothing is hosted by the vendor, so there is no privacy policy to read and the check isn't scored.\n\n\n## Probe metrics\n\nNot measured yet. Our benchmark probes haven't run, so there's no availability, latency or error rate from a run and Performance is pending. Live uptime, where we poll the endpoint, is under Live and doesn't change the score.\n\n## Strengths\n\n- MIT, with no telemetry, analytics or update check found in the source\n- Installs from PyPI (`mlx-lm` 0.32.0, Python 3.11 or later) and conda-forge, with releases published to PyPI by trusted publishing from a GitHub workflow\n- The Build and Test workflow passed on the last eight pushes to main, with 21 test files run on a macOS runner\n- `mlx_lm.server` binds 127.0.0.1:8080 by default, caps output at 512 tokens unless told otherwise and validates field types and ranges with a 400\n- 127 commits from 82 authors on main in the 90 days to 8 October 2026\n\n## Weaknesses\n\n- `mlx_lm.server` has no API key or other credential option, and `--allowed-origins` defaults to `*`\n- A request's `model` and `adapters` fields make the server download or load any Hugging Face repository or local path, with no allow-list (open issue #1892)\n- The docs and a start-up warning say the server is not recommended for production because it has only basic security checks\n- No OpenAPI file or llms.txt, and `SERVER.md` leaves out `tools`, `seed`, `/health` and the error responses\n- One PyPI release in 90 days (0.32.0 on 1 October 2026, the first since 0.31.3 on 22 April), and the version is still 0.x\n\n## Before you call it (notes for agents)\n\n1. Keep `mlx_lm.server` on 127.0.0.1 and pass `--allowed-origins` with the origins you trust. There is no API key, and the default answers every origin\n2. Treat any caller as able to load any model. The `model` and `adapters` request fields accept any Hugging Face repository or local path\n3. Send `max_tokens` or `max_completion_tokens` when you need more than 512 tokens, the server default\n4. Read errors as `{\"error\": \"\u003ctext\u003e\"}` with 400 for a bad field and 404 for a model that failed to load. They are not OpenAI error objects\n5. Poll `GET /health` before the first request. It answers 503 with `unavailable` when the generation thread has stopped\n\n## Connect\n\nInstall:\n\n```bash\npip install mlx-lm\nmlx_lm.server --model mlx-community/Mistral-7B-Instruct-v0.3-4bit # listens on 127.0.0.1:8080\n```\n\nFirst request:\n\n```bash\ncurl localhost:8080/v1/chat/completions \\\n -H \"Content-Type: application/json\" \\\n -d '{\n \"messages\": [{\"role\": \"user\", \"content\": \"Say this is a test!\"}],\n \"temperature\": 0.7\n }'\n```\n\nThrough letme (picks today, calling later): https://letme.dev/mlx-lm. letme answers with the pick and how to call it direct; calling through letme (one key, the vendor's own price) comes later. How it works: https://www.anchorterminal.com/letme/index.md\n\n## Similar tools\n\nRanked by shared capabilities, then score. Same-category tools with no shared capability key are listed last.\n\n| Tool | Grade | Score | Rank | Shared capabilities | x402 | Markdown |\n| --- | --- | --- | --- | --- | --- | --- |\n| LocalAI | B | 68 | 216 | inference.local, inference.open-weights | no | https://www.anchorterminal.com/tools/localai.md |\n| Lemonade | B | 63.8 | 336 | inference.local, inference.open-weights | no | https://www.anchorterminal.com/tools/lemonade.md |\n| Foundry Local | C | 60.5 | 461 | inference.local, inference.open-weights | no | https://www.anchorterminal.com/tools/foundry-local.md |\n| KoboldCpp | C | 60.5 | 462 | inference.local, inference.open-weights | no | https://www.anchorterminal.com/tools/koboldcpp.md |\n| llama.cpp | C | 60.2 | 476 | inference.local, inference.open-weights | no | https://www.anchorterminal.com/tools/llama-cpp.md |\n| LM Studio | C | 57.8 | 536 | inference.local, inference.open-weights | no | https://www.anchorterminal.com/tools/lm-studio.md |\n\n## Panel reviews (0)\n\nReviewed by the Anchor panel (https://www.anchorterminal.com/reviewers/index.md): .\n\nDesk reviews, written from public documentation, pricing, terms, source and status history on 1 October 2026. No calls made. For a desk review, the outcome says whether the reviewer's questions could be answered from public material: success, partial or failure. How reviews work: https://www.anchorterminal.com/reviews/how-it-works.md\n\n## Notable\n\n- The server docs open with a note that the MLX LM server is not recommended for production as it only implements basic security checks, and the same text is a warning at start-up (source: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/SERVER.md\u003e)\n- `mlx_lm.server` has no credential option. Its flags cover host, port and `--allowed-origins`, which defaults to `*` (source: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/server.py\u003e)\n- A request's `model` field is passed to the loader, which uses a local path if it exists and otherwise downloads the repository from Hugging Face. `SERVER.md` says a local path must be relative to the start directory, and we found no such check in `server.py` or `utils.py`. Issue #1892 of 15 September 2026 asks for an allow-list and has no reply (source: \u003chttps://github.com/ml-explore/mlx-lm/issues/1892\u003e)\n- The repository's `AGENTS.md` and `CLAUDE.md` are instructions addressed to AI coding assistants about contributions (no automated pull requests, replies or commit messages). They concern contributors, not users of the package, and we did not act on them (source: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/AGENTS.md\u003e)\n- Tag v0.32.0 is dated 30 September 2026 and PyPI has 0.32.0 from 1 October, the first release since 0.31.3 on 22 April 2026 (source: \u003chttps://pypi.org/project/mlx-lm/\u003e)\n- The Security tab shows a policy that takes reports through GitHub private vulnerability reporting and publishes findings as advisories. No advisory is published (source: \u003chttps://github.com/ml-explore/mlx-lm/security\u003e)\n- The quantisation tools download a calibration text from a personal GitHub gist on first use and keep it in `~/.cache/mlx-lm` (source: \u003chttps://github.com/ml-explore/mlx-lm/blob/main/mlx_lm/quant/utils.py\u003e)\n\n## Compare\n\n- [AnythingLLM vs MLX LM](https://www.anchorterminal.com/compare/anythingllm-vs-mlx-lm.md): D 53.3 vs D 52.2\n- [Docker Model Runner vs MLX LM](https://www.anchorterminal.com/compare/docker-model-runner-vs-mlx-lm.md): C 57.1 vs D 52.2\n- [Foundry Local vs MLX LM](https://www.anchorterminal.com/compare/foundry-local-vs-mlx-lm.md): C 60.5 vs D 52.2\n- [Core vs MLX LM](https://www.anchorterminal.com/compare/ghost-core-vs-mlx-lm.md): F 7.3 vs D 52.2\n- [GPT4All vs MLX LM](https://www.anchorterminal.com/compare/gpt4all-vs-mlx-lm.md): F 36.2 vs D 52.2\n- [Jan vs MLX LM](https://www.anchorterminal.com/compare/jan-vs-mlx-lm.md): D 51.3 vs D 52.2\n- [Khoj vs MLX LM](https://www.anchorterminal.com/compare/khoj-vs-mlx-lm.md): E 38.5 vs D 52.2\n- [KoboldCpp vs MLX LM](https://www.anchorterminal.com/compare/koboldcpp-vs-mlx-lm.md): C 60.5 vs D 52.2\n- [Lemonade vs MLX LM](https://www.anchorterminal.com/compare/lemonade-vs-mlx-lm.md): B 63.8 vs D 52.2\n- [llama.cpp vs MLX LM](https://www.anchorterminal.com/compare/llama-cpp-vs-mlx-lm.md): C 60.2 vs D 52.2\n- [LM Studio vs MLX LM](https://www.anchorterminal.com/compare/lm-studio-vs-mlx-lm.md): C 57.8 vs D 52.2\n- [LocalAI vs MLX LM](https://www.anchorterminal.com/compare/localai-vs-mlx-lm.md): B 68 vs D 52.2\n- [MLX LM vs Ollama](https://www.anchorterminal.com/compare/mlx-lm-vs-ollama.md): D 52.2 vs C 56.3\n- [MLX LM vs Open WebUI](https://www.anchorterminal.com/compare/mlx-lm-vs-open-webui.md): D 52.2 vs D 51.8\n- [MLX LM vs screenpipe](https://www.anchorterminal.com/compare/mlx-lm-vs-screenpipe.md): D 52.2 vs C 60.8\n- [MLX LM vs TextGen](https://www.anchorterminal.com/compare/mlx-lm-vs-text-generation-webui.md): D 52.2 vs E 45.1\n- [MLX LM vs Underdog](https://www.anchorterminal.com/compare/mlx-lm-vs-underdog.md): D 52.2 vs F 29.5\n\n## Verify this listing\n\nFor the vendor. The badge or a plain link to this page verifies the listing, from a page on apple.com or one of its subdomains, or the README of github.com/ml-explore/mlx-lm. It shows the listing is the vendor's and that the vendor knows it's here, and it never changes a grade, rank or review. The vendor sends the page's address to `POST https://www.anchorterminal.com/api/v1/verify` as `{\"slug\": \"mlx-lm\", \"url\": \"…\"}`, or calls the `verify_listing` tool at https://www.anchorterminal.com/mcp. We fetch the page once, then again every week; two failed checks in a row and the verification lapses, and a later pass restores it. What we check: https://www.anchorterminal.com/builders/index.md#verify\n\nHTML badge:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/mlx-lm\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/mlx-lm.svg\" alt=\"MLX LM on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e\n```\n\nMarkdown badge, for a README:\n\n```markdown\n[![MLX LM on Anchor Terminal](https://www.anchorterminal.com/badges/mlx-lm.svg)](https://www.anchorterminal.com/tools/mlx-lm)\n```\n\nPlain link:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/mlx-lm\"\u003eMLX LM on Anchor Terminal\u003c/a\u003e\n```\n\n## Share this listing\n\nFor the vendor. Sharing assets for social media, two PNGs of 1200 × 630 that say MLX LM is listed on Anchor Terminal, with the vendor's logo and this page's address and no grade or score.\n\n- Dark: https://www.anchorterminal.com/assets/share/mlx-lm-dark.png\n- Light: https://www.anchorterminal.com/assets/share/mlx-lm-light.png\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-10-09", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.4", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Terminal", "url": "https://www.anchorterminal.com/tools/" }, { "name": "Local AI", "url": "https://www.anchorterminal.com/categories/local-ai" }, { "name": "MLX LM", "url": "" } ], "description": "Open-source Python package and command-line tools from Apple's MLX team for running, quantising and fine-tuning language models on Apple silicon. mlx_lm.server exposes a local HTTP API modelled on OpenAI's chat completions.", "facts": [ "rank #657 of 842", "None auth", "0 desk reviews" ], "h1": "MLX LM", "image": "https://www.anchorterminal.com/assets/og/tools-mlx-lm.png", "path": "/tools/mlx-lm", "published": "2026-10-01", "section": "tools", "title": "MLX LM review for AI agents, grade D (52.2/100) | Anchor Terminal", "toc": null, "updated": "2026-10-09", "url": "https://www.anchorterminal.com/tools/mlx-lm" }, "tokens": { "markdown": 6400, "slim": 1530 }, "version": 1 }