AGENT INTEGRATION

Connect your Agent. Keep domain knowledge current.

The platform provides the knowledge runtime. Your Agent supplies its model, source access, external tools and scheduling.

MCP

The console provides remote MCP settings for each space. Read connections expose 8 reading tools; maintenance connections add 7 maintenance tools for 15 in total. Server-side credentials determine actual access.

Read & understand

get_context · search_knowledge · get_graph · explain_paths · get_evidence · read_source · list_skills · get_skill

Model & maintain

check_ontology · put_source_snapshot · compare_sources · submit_change · get_change · record_run · save_skill

Maintenance flow

Give your Agent the operating guide so it preserves evidence, revisions and capability boundaries. Read the Agent operating guide

  1. Read context, including the model, current revisions and capabilities.
  2. Create a run and claim its lease. Maintain checkpoints while processing sources.
  3. Save source snapshots. Reference returned snapshot IDs in model or knowledge changes.
  4. Read receipts and confirm application. Changes requiring confirmation are handled by the user in the console.
  5. Complete the run with all its applied change receipts.

HTTP API

Use the same Agent credential. Writes use pinned revisions, idempotency and atomic transactions. The connection panel provides actual endpoints for the current space.

Authorization: Bearer YOUR_AGENT_TOKEN

GET  /v2/spaces/{space_id}/context
POST /v2/spaces/{space_id}/changes
POST /v2/spaces/{space_id}/query/evidence

Capability boundaries

Agents cannot manage accounts, subscriptions, space closure or platform operations. Model upgrades and identity resolution need explicit grants. Breaking model changes require user confirmation. Pausing a run invalidates its lease; it does not stop the Agent process on your machine.

What can a customer Agent do?

A customer Agent uses a space-scoped credential to read context, query the graph, inspect evidence and submit maintenance changes within its grant. Higher-risk actions such as model upgrades, identity resolution and candidate Skill activation remain behind space capability boundaries.

Does the user question language affect persisted knowledge?

Yes. The Agent should use the language of the latest user request as the generation baseline while preserving the original language of sources and evidence. Machine identifiers stay stable when the interface language changes.