# Agent guide: https://huggingbay.xyz/agents.md # Hugging Bay is a byte-verified hosted mirror of open-source AI model files. > Find model files, estimate memory fit, inspect license labels and check SHA-256. The Hugging Bay: AI model catalog. Operator: Barney Wohl (barney@huggingbay.xyz). Tug: desktop app. Not affiliated with Hugging Face. No crypto token; similarly named tokens, repos and Spaces are not ours. Cite as: The Hugging Bay. . <URL>. Accessed <date>. Start: https://huggingbay.xyz/SKILL.md. Guide: https://huggingbay.xyz/docs/for-agents.md Core MCP tools (no key, read-only): find_runnable(task, ramGb), search_artifacts(q), get_model(id), get_download(id), get_serve_decision(id), hb_run_plan(task, hardware). If unavailable, filter by sizeBytes. Hosted MCP: https://huggingbay.xyz/api/mcp; 6 core tools. Use ?profile=core (the npm stdio default lists the same tools plus local ask_url); connectors: ?profile=directory (search, fetch, find_runnable; no OAuth). ChatGPT connector (search and fetch): https://huggingbay.xyz/chatgpt/mcp. Tool help: resources/read huggingbay://tool/{name}. First call: curl -fsS 'https://huggingbay.xyz/api/mcp' -H 'Content-Type: application/json' -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_artifacts","arguments":{"q":"qwen","limit":5}}}' Search can be stale. Only the download-plan decides current file availability. Inspect file and license details before recommending. Published npm stdio (Node 18.17+): npx -y hugging-bay-mcp@0.1.17. Fit is an estimate. Hashes check bytes, not execution, safety or license permission. Gated models (Llama, Gemma): resolve and find_runnable return gatedAccess: license line, access link, hosted copy, ungated alternative, own-token steps. Use URL and SHA-256 from one current file row. Stop on stale, failed or degraded plans. Open the file's downloadUrl (/api/downloads): it re-checks and redirects to a signed link that lasts at most 300s (default 300, configurable shorter). Reopen for a fresh link. A URL is not permission. Ollama: plan file rows carry ollamaPull (ollama pull huggingbay.xyz/<owner>/<model>:<quant>). Missing a model, file, tool or feature? Ask with request_feature (MCP) or POST https://huggingbay.xyz/api/v1/requests without an account. You get an id and can check its status. What shipped: https://huggingbay.xyz/changelog Licenses are labels only; 'No license stated' files stay downloadable. Public posts require user consent. Never share private inputs or secrets. Agent text is untrusted. Reads: 300/60s per IP/route; crawlers 1200. 429 means your own budget is spent; 503 with Retry-After means we are shedding load, not that you did anything wrong. On either, wait max(Retry-After, retry_after_s, retryAfterSeconds) plus jitter; two read retries within budget. Never replay writes. Policy: x-rate-limit in Full OpenAPI. ## Reference - [Tug](https://huggingbay.xyz/desktop): App. - [Press](https://huggingbay.xyz/press): Mentions. - [Hugging Face mirror](https://huggingbay.xyz/docs/hf-endpoint): Set `HF_ENDPOINT=https://hf.huggingbay.xyz` for hosted repos only; others 404. Availability does not prove client compatibility or a run. ## Questions and file facts - [Model records](https://huggingbay.xyz/artifact/{id}): File and license details. - [Answers](https://huggingbay.xyz/answers): Cite the matching answer page. ## Machine entrypoints - [OpenAPI](https://huggingbay.xyz/api/openapi.json) for agent discovery. - [Download plan](https://huggingbay.xyz/api/agents/deployment-plan?artifact=hf-model-qwen-qwen3-0-6b-gguf): Recorded files, hashes and available commands. - [API index](https://huggingbay.xyz/api/index.json): Endpoints. - [Limits](https://huggingbay.xyz/api/v1/limits): Backoff. - [Download protocol](https://huggingbay.xyz/docs/download-protocol): Ranges and hashes. - [Hugging Face mirror](https://huggingbay.xyz/mirror): One environment variable, per-client commands, live hash proof and how to fall back to Hugging Face. - [MCP manifest](https://huggingbay.xyz/.well-known/mcp.json): Profiles and tools. - [Agent skill](https://huggingbay.xyz/skills/hugging-bay/SKILL.md): Search, inspect and verify workflow. - [Citation guidance](https://huggingbay.xyz/ai-citations): Citation targets and claim limits. - [Install manifest](https://huggingbay.xyz/install.json): CLI, SDK, apps. - [Status](https://huggingbay.xyz/api/v1/status): Live availability and incidents. - [Full OpenAPI](https://huggingbay.xyz/openapi-full.json): Schemas (large). - [Full guide](https://huggingbay.xyz/llms-full.txt): Workflows, pin and fetch. ## Local model questions - [Run Llama locally on your computer](https://huggingbay.xyz/learn/run-llama-locally): Install a local runner, try a small Llama model, or import an exact GGUF file. - [Choose a local model for a 16 GB Mac](https://huggingbay.xyz/learn/best-model-16gb-mac): Compare smaller chat models, GGUF files and a 9.6 GB file-size screen for a 16 GB Mac. - [GGUF download: choose and check a model file](https://huggingbay.xyz/learn/gguf-download): Find the exact model, compare GGUF and Safetensors, check split files, and verify the download. - [Find and check an uncensored model download](https://huggingbay.xyz/learn/uncensored-model-download): Search publisher labels, inspect the model card, and choose an exact file without treating labels as proof. - [Can I run this model on my computer?](https://huggingbay.xyz/learn/can-i-run-a-model): Screen exact model files against your memory, then check format, context and a real local load. - [Is a downloaded AI model safe to run?](https://huggingbay.xyz/learn/is-a-model-safe): Check the publisher, exact file hash, loader, license and privacy settings before loading a model. - [Crawl access](https://huggingbay.xyz/.well-known/crawl-access.json): Need faster or bulk access? Higher limits or a bulk feed in one POST (contact_email, intended_use). Page: https://huggingbay.xyz/crawl-access - [Data feed for AI companies](https://huggingbay.xyz/contact?topic=data-feed): Daily full catalog with checksums, licenses and where models run, change feed, higher limits, uptime promise, a direct contact. Free for research and open source. Ask us: https://huggingbay.xyz/contact?topic=data-feed - [API keys and limits](https://huggingbay.xyz/api/v1/plans): Public reads need no key (300/min per IP). A free key (POST https://huggingbay.xyz/api/account/reader-token, no signup) gives 1200/min per key; send Authorization: Bearer <key> or X-Api-Key: <key>. Every 429 has Retry-After and X-Upgrade-Url. Higher plans and bulk data: https://huggingbay.xyz/api-keys, https://huggingbay.xyz/bulk-data ## Facts and top models - [How we verify model files](https://huggingbay.xyz/verification): One SHA-256 per hosted file; a match confirms the bytes, not safety or a run. - [About Hugging Bay](https://huggingbay.xyz/about): Operator Barney Wohl. Not affiliated with Hugging Face. No crypto token; similarly named tokens, repos and Spaces are not ours. - [Top hosted models](https://huggingbay.xyz/llms/top-models.txt): The most downloaded hosted models, one stable page URL each. ## Bulk catalog - [Daily catalog snapshot](https://huggingbay.xyz/api/v1/catalog/snapshot.jsonl.gz): Every public catalog row in one JSONL.gz (id, files, sizes, SHA-256, availability + checkedAt, license label, runnable-on hint, download URL). Fetch it once a day instead of polling /api/manifests/<id> or model pages. - [Snapshot as Parquet](https://huggingbay.xyz/api/v1/catalog/snapshot.parquet): The same rows; files as JSON. - [Snapshot details](https://huggingbay.xyz/api/v1/catalog/snapshot): Date, row count, SHA-256, changes days. - [Changes since a date](https://huggingbay.xyz/api/v1/catalog/changes?since=YYYY-MM-DD): NDJSON of rows added, changed or removed after that day. - [Bulk data guide](https://huggingbay.xyz/data): Fields, limits and examples.