AI agents

AI execution is never synchronous HTTP between NestJS and FastAPI. The API enqueues jobs; the FastAPI worker consumes them and publishes results.

Registered agents

DB key (`AgentDefinition`)	Worker router	Mode	Processor
`requirements`	`requirements`	Conversational	`requirements_processor.py`
`developer_intelligence`	`developer_intelligence`	Conversational	(graph node)
`project_planner`	`planner`	One-shot	`planner_processor.py`
`qa_intelligence`	`qa_intelligence`	Conversational	`qa_intelligence_processor.py`
`change_impact`	— (planned)	Conversational	Not in router yet

LangGraph entry: app/core/graph/router.py — conditional route on agent_type in job payload.

Queue contract

Prop

Type

Backend publishes after REST handlers return 202 Accepted. NestJS reads results and updates ProjectAiChatMessage, RagnarockChatMessage, tasks, etc.

NestJS → agent triggers

User action	API endpoint	Agent
Requirements chat turn	`POST .../ai/requirements`	`requirements`
File upload	`POST .../ai/requirements/upload`	`requirements`
Ragnarock message	`POST .../ai/ragnarock/chat`	Router picks agent
Architecture doc	`POST .../ai/arch-doc/generate`	`developer_intelligence`
Plan	`POST .../ai/plan/generate`	`planner`
QA suite	`POST .../ai/qa/generate`	`qa_intelligence`

Vector memory

Optional pgvector database (:5433 in dev compose) stores embeddings for past SRS content and Q&A. Modules: app/memory/retrieval.py, vector_store.py.

LLM providers

Set in ragnarock-agent/.env:

Gemini (default path in many deployments)
Anthropic
OpenAI

Factory: app/core/llm/factory.py reads LLM_PROVIDER.

Run locally

cd ragnarock-agent
cp .env.example .env
pip install -r requirements.txt
./run.sh   # listens :8100
curl http://localhost:8100/health

Docker: docker/staging/Dockerfile.agent in backend repo.

Product docs

Ragnarock AI — user-facing chat
Documentation workspace — where SRS output lands
Test Cases — QA agent output