Skip to main content
ReferenceMCP toolDraft · not yet verified

e_ingest

Store new knowledge — one fact per SR.

Agent Guidance​

This is the description the engine gives an agent for this tool, word for word.

Store new knowledge — one fact per SR. Content should stand alone (include the reasoning/context, not just the conclusion).

Key inputs:

  • source_type: conversational (default) | documentary — validated against the profile/lane's allowed types (an operator may extend the allowlist or allow any).
  • properties: ["label:value", ...] — dimensions you'd filter or compare on (not content). Reserved e_* labels are rejected.
  • is_conversational: opt the SR into the conversational continuum — enables window-expansion / neighbor recall so adjacent turns are retrieved together. Set true for dialogue you want recalled in context (e.g., a user/assistant exchange whose surrounding turns matter).
  • is_file_content: marks the SR as a contiguous chunk of a larger file (boosts windowed recall across that file's parts). Leave false unless ingesting file parts via the file-upload skill/script.
  • assign_lane: send the SR to a specific lane you're authorized to write; omit to use your profile's home lane. Name it by its name, by its qualified name name@fragment (shown where two lanes you can see share a name), or by its lane uuid; an ambiguous name is refused with the qualified names to choose from.
  • adjusts: [sr_ids] (+ optional adjustment_label) — link this new SR to prior SR(s) it refines, clarifies, or extends (knowledge accumulating — not error; use e_correct for wrong). Usually 1:1 (an adjustment edge between two SRs); name several to bind a tight group under one shared token. The token is stamped on the new SR and each writable target (others skipped + reported), shown on e_read and walkable by e_gather traverse.

Returns the new sr_id (plus an adjustment summary when adjusts is used).

Arguments​

ArgumentTypeDefault
contentstringrequired
source_typestringconversational
is_conversationalbooleanfalse
is_file_contentbooleanfalse
propertieslist of strings—
assign_lanestring—
adjustslist of strings—
adjustment_labelstring—
agent_profilestring—

Refusals​

A refusal is returned as {"error": "…"}. These are the ones this tool raises itself; … stands for a value filled in at the time.

Text
content is required and cannot be empty or whitespace-only.
lane '…' is declared by no policy this engine runs; writable: …
labels … are reserved (the 'e_' namespace is engine-minted). To link an updating SR to existing ones, pass adjusts=[sr_id, ...].
agent_profile cannot write to lane '…'; writable: …
agent_profile has no home_lane
…; writable: …