Kinic Wiki

Read task-scoped wiki context through query_context. Use this first for normal questions and before broad source investigation; answer from returned nodes and evidence, not search hits alone.

Fetch full text for up to 10 Kinic Wiki search results. Pass each result's exact id or public url from search. Never pass ChatGPT citation tokens such as turn0file0. If ChatGPT hides an id behind a citation token, construct the public url under the configured Kinic Wiki public origin as /db/{database_id}{path} from that result's metadata. Use after candidate search for broad or list questions, including for a single strongest result. Item-level errors are returned without failing the whole call.

Use this when you need to discover which Kinic Wiki database visible to the current caller matches a user task.

List nodes under a prefix without reading content. Use prefix / for root inventory and prefix discovery, /Knowledge for curated notes, and /Sources for raw evidence inventory.

Discover Store API roots, capabilities, roles, and limits for one Kinic Wiki database visible to the current caller.

Preferred for the whole request when any operation is append, edit, multi-edit, mkdir, move, or delete, including one operation. Apply 1 to 100 ordered write, append, edit, multi-edit, mkdir, move, and delete operations atomically in one database; any failure rolls back the whole batch. Changes to a public database or published node can become publicly visible.

Read full text for one known VFS path without a search result id. Use when list, context, or prior evidence already supplied the path.

Read full text for 2 to 10 known VFS paths in one restricted SQL query. Use for multiple paths from list, context, or search metadata instead of repeated read_path/fetch calls.

Search one selected Kinic Wiki database visible to the current caller. Use multiple queries for broad/list/classification tasks. Use prefix / for whole-DB recall, /Knowledge for curated notes, and /Sources for raw evidence when curated notes are thin. Use content-start preview for candidate classification.

Preferred for every create/full-replacement request, including one node. Create or replace 1 to 100 nodes in order and atomically in one database; any failure rolls back the whole batch. Changes to a public database or published node can become publicly visible.