neosian docs

Quickstart Install, boot keylessly, run an agent, wrap it in a Conversation Agent AgentConfig field by field, the client seam, hooks vs wire events, the knobs Local serving A door that signs nothing, the zero card, the two recipes and the measured local row Tools @Tool: what the signature becomes on the wire, what is validated, what each provider keeps Memory Small documents, mounts and scopes, six commands, every write versioned Skills a skill is skills/<name> in a mount - versioned, curated by flag, served as MCP prompts Stores The two ABCs, the seven constraints, the kits as the standard, a SQLite walk, the table The shell status, setup, chat, search, audit, eval; the memory grammar, --json, exit tiers MCP memory, skills, recall and search over stdio; mcp install; McpServer consumes any server Any agent Agent hooks record sessions; startup context, MCP recall and search read them back Interop The memory tool under pydantic-ai, the Agents SDK or an MCP client; the data stays yours Topology, not hierarchy Who runs neosian code, where the bytes live, and one writer per root The wire The nineteen /v1/ routes, the envelope, paging, the JSON shapes and WIRE_VERSION Memory baselines Self-measured memory-layer results per provider and transport, fingerprint-gated

Nothing is authored on this site: a correction is a pull request to mausa-ai/neosian. Agents start at llms.txt, the machine-readable front door (byte-identical to the wheel’s copy). The library is on PyPI and GitHub; neosian.com is the front door for people.