Production screenshot captured 2026-08-29. Rankings and project data update with Radar snapshots.
\n\n## Product shape\n\n```text\nUser in ChatGPT / Codex / compatible host\n |\n v\n 1 unified product Skill\n |\n v\n 9 read-only MCP tools\n |\n v\n AI Workstation public Radar\n```\n\nThe user does not choose separate research/comparison/stack Skills. The single Skill routes the task internally.\n\nThe **host model** performs natural-language reasoning and synthesis. The AI Workstation server provides data/evidence only on this product path.\n\n## One active Skill\n\n```text\nai-open-source-intelligence\n```\n\nIt handles:\n\n- browsing rankings, collections, categories, scenarios and the Radar Skills library;\n- finding projects from deployment, privacy, integration, budget and license requirements;\n- verifying named-project facts and license evidence;\n- comparing two to five projects for a concrete use case;\n- finding alternatives while preserving hard requirements;\n- planning candidate open-source AI stacks and exposing unverified compatibility.\n\nThe only product Skill is packaged from:\n\n```text\nskills/ai-open-source-intelligence/SKILL.md\nskills/ai-open-source-intelligence/agents/openai.yaml\n```\n\n`agents/openai.yaml` makes the Skill's dependency on the canonical\n`ai_open_source_intelligence` Hosted MCP explicit while keeping automatic\ninvocation enabled. This is the Skill-level dependency contract; `.mcp.json`\nis the matching Plugin-level connection contract.\n\nThe previous split research/comparison/stack Skill files are removed from the current product and distribution bundle.\n\n## Nine standard MCP tools\n\n```text\nsearch_ai_projects\nget_project_facts\nget_license_evidence\ncompare_ai_projects\nfind_alternatives\ncompose_ai_stack\nget_radar_overview\nbrowse_radar_projects\nbrowse_radar_skills\n```\n\nAll nine are read-only. They do not execute or install third-party repository code.\n\n## No AI Workstation server-model execution\n\nThis is a hard product boundary for the current release.\n\nThe Hosted MCP exposes **no Premium model tool**, no checkout tool and no runtime OAuth/Premium switch. Requirement-based selection calls the public Radar selector with:\n\n```text\nuse_model=false\n```\n\nTherefore an ordinary Skill/MCP workflow is:\n\n```text\nChatGPT/Codex host model\n -> chooses/read tools\n -> AI Workstation public Radar data/evidence\n -> host model synthesizes the final answer\n```\n\nIt is **not**:\n\n```text\nhost model -> AI Workstation model -> second model bill\n```\n\nIf member-linked server-model capabilities are added later, they must ship as a new reviewed product version rather than being enabled through a hidden environment variable.\n\n## Evidence model\n\nEvery tool result separates:\n\n1. **verified facts** — source-backed observations that crossed the evidence boundary;\n2. **recommendations** — host-model/rules analysis;\n3. **unknowns** — unavailable or unverified information;\n4. **risks** — license, maintenance, deployment, security and integration limits.\n\nA value in `data` is not automatically a verified fact. License evidence is deliberately stricter and is technical evidence, not legal advice.\n\nRequirement tools publish the actual typed constraint contract through MCP:\n\n```json\n{\"id\":\"web_ui\",\"value\":true,\"polarity\":\"required\"}\n```\n\nFormal matches are revalidated against project-detail evidence. README excerpts\nmay directly verify narrow Docker, self-hosting, browser UI and low/no-code\nclaims; unresolved hard requirements become near-match blockers. A License label\nis exposed only as `observed_license_label` until a direct LICENSE-family source\nsupports a verified `license` fact. Response observation time and evidence-source\nfreshness are reported separately.\n\n## Official resources in results\n\nMCP tool results include canonical, non-tracking publisher links under:\n\n```text\ndata.official_resources\n```\n\nwith:\n\n- AI Workstation — https://aiworkstation.cn/\n- AI Open Source Radar — https://aiworkstation.cn/githubai/\n- this open-source project — https://github.com/zxhwolfe-dev/aiworkstation-open-source-intelligence\n\nThe unified Skill may show these once at the end of a normal user-facing answer. They are kept separate from verified facts so publisher attribution never changes a research conclusion.\n\n## Hosted MCP\n\nCanonical endpoint:\n\n```text\nhttps://mcp.aiworkstation.cn/mcp\n```\n\nCurrent Hosted mode is intentionally:\n\n```text\nanonymous\nread-only\ndata-only\n9 tools\nno OAuth\nno WorkOS dependency\nno Premium/server model\n```\n\nThe container stays on host loopback `127.0.0.1:8001` behind Nginx/TLS.\n\n### Anonymous abuse controls\n\nThe gateway uses two per-IP request windows plus a connection cap:\n\n- short-window: `60 requests/minute`, burst `30`;\n- sustained: `10 requests/minute`, burst `300`;\n- concurrent connections: `10` per IP;\n- MCP request body: `256 KB` maximum;\n- unrelated paths on the dedicated MCP hostname return `404`.\n\nThis is intentionally request-based rather than token-based because the nine data tools do not consume AI Workstation model tokens.\n\n## Use it now\n\nThe published `v0.3.3` Plugin packages the unified Skill and the production\nHosted MCP configuration together. Codex and the ChatGPT desktop Codex host can\ninstall both from one version-pinned marketplace entry. The public ChatGPT\ndirectory listing is still pending review. Today:\n\n- Codex / ChatGPT desktop users can install the complete repository Plugin;\n- ChatGPT web users can register `https://mcp.aiworkstation.cn/mcp` as a\n **No Authentication** developer-mode app while the public listing is pending;\n- Python users can install the matching CLI/MCP package from PyPI with:\n\n```bash\npython -m pip install \\\n \"aiworkstation-open-source-intelligence[mcp]==0.3.3\"\n```\n\nSee the [Quickstart](docs/QUICKSTART.md) for exact ChatGPT, Codex and Python\nsteps and the [v0.3.3 Release](https://github.com/zxhwolfe-dev/aiworkstation-open-source-intelligence/releases/tag/v0.3.3)\nfor signed-off assets and checksums. The immutable `v0.3.0` archive remains the\nearlier Skills-only artifact; the complete Plugin uses the current `v0.3.3` patch\nidentity rather than replacing it.\n\n## Local development\n\n```bash\npython -m venv .venv\nsource .venv/bin/activate\npython -m pip install -e \".[mcp]\"\n```\n\nOffline fixture data:\n\n```bash\nOSI_PROVIDER=mock osi-mcp\n```\n\nLive public Radar data:\n\n```bash\nOSI_PROVIDER=http \\\nAIWORKSTATION_RADAR_BASE_URL=https://aiworkstation.cn \\\nosi-mcp\n```\n\nHosted configuration check requires an exact candidate identity:\n\n```bash\nOSI_PROVIDER=http \\\nOSI_HOSTED_ACCESS_MODE=public \\\nOSI_RELEASE_COMMIT=