ActiveBlue AI

Multi-agent AI system integrated with Odoo 18 Community Edition, powered by Ollama (activeblue-chat / llama-based model) running fully on-premise.

What's New (May 2026)

• Ollama cold-start fix: activeblue-chat model takes ~124s to load from disk; timeout raised to 300s (enforced in code regardless of env var) and the model is pre-warmed at service startup so the first user message never times out.
• Model pre-warming: _prewarm_ollama() is launched as a background task during lifespan startup, loading the model into VRAM before any user traffic arrives.
• Improved health check: /health/detailed now queries Ollama's ps endpoint to report whether the model is loaded (ok) or still loading (warming), giving accurate bot online/offline status.
• Comprehensive unit test suite: 433 tests across all 8 specialist agents, all tool layers, PeerBus, AgentRegistry, ToolCallValidator, and base agent lifecycle. Run with .venv-test/bin/python -m pytest tests/ -q.
• Tool count enforcement: Each specialist agent is validated at startup to have ≤ 8 tools (AgentConfigError otherwise).
• PeerBus inter-agent communication: Agents can call each other with depth-limited routing, timeout safety, and call-log tracking.
• Auto-RAG: All agents automatically fetch Odoo 18 workflow guidance from odoo_doc_agent before answering.
• Auto-heal loop: Background task calls sysops_agent.auto_heal() every 2 minutes if any system is degraded.

Architecture

Odoo 18 (ai.activeblue.net)
  └── activeblue_ai module
        ├── OWL2 systray brain icon + slide-in chat panel
        ├── Models: ab.ai.bot, ab.ai.directive, ab.ai.log, ab.ai.agent.registry
        └── Controllers: /ai/chat, /ai/webhook/callback, /ai/health, /ai/approval/*

FastAPI Agent Service (192.168.2.47:8001)
  ├── POST /dispatch       — route user message to MasterAgent
  ├── GET/POST /approval/* — human approval workflow
  ├── GET/POST /registry/* — agent registry + LLM backend overrides
  ├── POST /sweep          — trigger proactive agent sweeps
  └── GET /health          — service health + Odoo/Ollama status

MasterAgent (singleton)
  ├── Classifies intent via LLM
  ├── Routes to specialist agents in parallel (asyncio.gather)
  ├── Manages 3-tier memory (conversation / operational / knowledge)
  └── Synthesises responses

Specialist Agents (8, stateless):
  finance_agent, accounting_agent, crm_agent, sales_agent,
  project_agent, elearning_agent, expenses_agent, employees_agent

Privacy Modes

Mode	Behaviour
local	Ollama only for all agents
hybrid	Per-agent override (DB → env → fallback)
cloud	Claude for non-HIPAA agents

HIPAA-locked agents (always Ollama, no exceptions): finance_agent, accounting_agent, employees_agent, expenses_agent

Quick Start

1. Clone and configure

git clone http://192.168.1.64:3000/tocmo0nlord/odoo-ai.git
cd odoo-ai
cp .env.example .env
# Edit .env — set POSTGRES_PASSWORD, ODOO_API_KEY, etc.

2. Run Odoo 18

docker compose -f docker-compose.odoo.yml up -d

3. Run the Agent Service

docker compose up -d

Or for development:

pip install -r requirements.txt
uvicorn agent_service.main:app --reload --port 8001

4. Run database migrations

cd agent_service/migrations
alembic upgrade head

5. Install Odoo module

In Odoo → Settings → Apps → search "ActiveBlue AI" → Install.

Environment Variables

See .env.example for the full list. Key variables:

Variable	Description
ODOO_URL	Odoo base URL (e.g. http://ai.activeblue.net)
ODOO_API_KEY	Odoo user API key
OLLAMA_URL	Ollama API URL (e.g. http://192.168.2.47:11434)
ANTHROPIC_API_KEY	Required only if LLM_PRIVACY_MODE=cloud or hybrid
LLM_PRIVACY_MODE	local / hybrid / cloud (default: local)
POSTGRES_PASSWORD	Required — no default
WEBHOOK_SECRET	Shared secret between Odoo and agent service

Development

Running tests

# Using the project test venv (recommended)
.venv-test/bin/python -m pytest tests/ -q

# Or install manually
pip install pytest pytest-asyncio
pytest tests/ -v

Test coverage (433 passing, all on Ollama/local mode):

• tests/test_registry.py — AgentRegistry lifecycle
• tests/test_peer_bus.py — PeerBus routing, depth limits, timeouts
• tests/test_tool_validator.py — ToolCallValidator, type coercion, enum guards
• tests/test_*_tools.py — 8 files covering every tool method (finance, accounting, crm, sales, project, elearning, employees, expenses)
• tests/test_*_agent.py — 8 files covering plan/gather/reason/act/report/sweep/peer_bus for each specialist agent
• tests/test_dispatch_router.py, test_e2e_dispatch.py, test_llm_router.py, test_odoo_client.py — integration tests

Project structure

odoo-ai/
├── agent_service/
│   ├── agents/          # MasterAgent + 8 specialist agents + PeerBus + SweepCoordinator
│   ├── llm/             # OllamaBackend, ClaudeBackend, LLMRouter, ToolCallValidator
│   ├── memory/          # ConversationStore, OperationalStore, KnowledgeStore, MemoryManager
│   ├── tools/           # OdooClient + per-domain tools (finance, crm, sales, ...)
│   ├── routers/         # FastAPI routers (dispatch, approval, registry, sweep, health)
│   ├── prompts/         # System prompts for each agent
│   ├── migrations/      # Alembic migrations (7 tables)
│   ├── logging_utils/   # Structured JSON logging + Loki push
│   ├── config.py        # pydantic-settings
│   ├── app_state.py     # Global singletons
│   └── main.py          # FastAPI app + lifespan startup
├── addons/
│   └── activeblue_ai/   # Odoo 18 module
│       ├── models/      # ab.ai.bot, ab.ai.directive, ab.ai.log, ab.ai.agent.registry
│       ├── controllers/ # webhook, health_proxy, approval + chat
│       ├── views/       # XML views + menus
│       ├── security/    # groups + ACL
│       ├── data/        # cron jobs
│       └── static/      # OWL2 JS + CSS + XML templates
├── research/            # Per-domain research notes
├── tests/               # pytest test suite
├── docker-compose.odoo.yml  # Odoo 18 + PostgreSQL 15
├── docker-compose.yml       # Agent service + PostgreSQL 15
├── Dockerfile
├── requirements.txt
└── .env.example

Agent Tool Limits

Each specialist agent is capped at 8 tools (MAX_TOOLS_PER_AGENT). The ToolCallValidator raises AgentConfigError at startup if exceeded.

Memory Architecture

Tier	Store	TTL	Scope
Tier 1	ab_conversation_memory	Hard cap: 200 rows/user	Per user
Tier 2	ab_operational_memory	90 days	Per agent+scope
Tier 3	ab_knowledge_store	Permanent	Entity-keyed

License

LGPL-3.0