Docs index
Use your results with AI
Choose live MCP/API access or the signed portable package, and keep answers grounded in the same evidence.
Live access or a portable package
There are two supported ways to use TAVONEL output. Live access through Ask, the API or the read-only MCP server reads the active World and is the preferred path for a production assistant that needs the current revision. The signed knowledge package is a portable snapshot for offline work, handoff, archive and systems that consume files rather than an API.
Which surface for which use
| Use case | Recommended surface | Why |
|---|---|---|
| AI/agent that should stay current | MCP or API | Reads the active World, access policy and evidence without copying a stale snapshot. |
| TAVONEL workspace Q&A | Ask | Uses the same active World and returns source-grounded citations. |
| Local coding/research agent with folder access | Signed ZIP | Extract once, grant the agent folder access, and let AGENTS.md describe the package contract. |
| Web chat with file upload but no local filesystem | Signed ZIP or selected package files | The chat must receive the bytes; a local path by itself does not grant access. |
| Graph/RDF/RAG import | Signed ZIP | Use the projection that matches the target system while retaining validation and provenance beside it. |
Setting either path up
- Before using a result as organizational truth, confirm that the World you intend to use is active. A review_required candidate is not automatically authoritative.
- For a live agent, create a least-privilege API key and use the read-only MCP bridge or the REST API from the Developers surface.
- For a portable workflow, download the signed knowledge package and extract it without changing its internal paths.
- Give a filesystem-capable agent access to the extracted folder and tell it to read AGENTS.md first. The machine-readable manifest/ai-entrypoint.json points to retrieval, ontology, graph, provenance and validation files.
- Require the consuming AI to preserve uncertainty and cite the package evidence/source locator it actually used. README.md and AGENTS.md are instructions, not evidence sources.
Telling an agent what to read first
Read AGENTS.md in this folder first. Use manifest/ai-entrypoint.json to locate the compiled knowledge and evidence. Answer from this package, preserve uncertainty, and cite the source evidence you relied on. If the task requires the latest organizational state, tell me to use the live TAVONEL MCP/API instead of assuming this snapshot is current.What each package file is for
| Package path | Use it for |
|---|---|
| README.md | Human-facing start guide and integration choices. |
| AGENTS.md | Instructions for a filesystem-capable AI/agent. |
| manifest/ai-entrypoint.json | Machine-readable map of the portable entrypoints and grounding rules. |
| rag/chunks.jsonl | Generic retrieval/RAG ingestion. |
| rag/documents.jsonl | Document-level retrieval records. |
| ontology/knowledge.jsonld | JSON-LD semantic/ontology consumers. |
| ontology/knowledge.ttl | RDF/Turtle graph consumers. |
| graph/nodes.csv + graph/relationships.csv | Simple graph imports. |
| provenance/activities.jsonl | Lineage and provenance inspection. |
| validation/report.json | Whether the result passed validation or still requires review. |
| manifest/export-manifest.json + signatures/ | Integrity verification for every signed package entry. |
Note A filesystem path is not a connector. If an AI application cannot read local files, upload the package or use MCP/API instead of pasting a path it cannot access.
API version 2026-09-02.1 · reviewed 11 September 2026
Something here out of date or wrong? Report an issue with this page.