LLM Enrichment and AI Analysis
Providers (priority order): Ollama (1st, free), Anthropic (2nd), OpenAI (3rd). 22 handler enrichments. Security scan analysis (risk summary, root causes, remediation). Token tracking per interaction.
12.1 Enrichment Microservice (Optional Companion)
Section titled “12.1 Enrichment Microservice (Optional Companion)”For deployments that need agent-builder workflows separate from the main gateway, AgentWatch ships an optional companion microservice in enrichment-service/ — a Bun-based Express service (default port 3101) that runs alongside the main server.
What it provides:
- Agents API — CRUD for agent definitions, with prompts, strategies, knowledge sources, and tool bindings
- Connectors — pluggable integrations (file system, web fetch, custom HTTP endpoints)
- LLM service — multi-provider chat with retry, failover, and per-tool routing
- Tools API — agent-callable tools registered as JSON schemas
- Workflows — multi-step agent execution graphs
- MCP knowledge sources — distinct from MCP tools: MCP servers can be registered as knowledge sources that supply contextual data via MCP Resources (not just callable tools). The
injectKnowledge()pipeline fetches resources from configured MCP servers and injects them into the agent’s system prompt before execution - UI assets — bundled agent-builder UI
This service is independent of the main AgentWatch gateway — customers can deploy it (Docker + dedicated Postgres or shared) only if they need the agent-builder surface. Default deployments don’t need it.