Fred

Fred, released by Thales Group under the Thulite brand, is a full-stack runtime platform for multi-agent applications. It integrates four core components: Agentic Backend (multi-agent runtime and session orchestration), Knowledge Flow Backend (document ingestion and RAG retrieval), Control Plane Backend (team management and ReBAC access control), and a React frontend.

The platform provides a v2 Agent SDK supporting two authoring styles: ReAct/Profile Agents (declarative roles and tools with SDK-managed execution loops) and Graph Agents (explicit state management, conditional routing, and human-in-the-loop approval gates). Both styles support MCP tool integration and run within the same runtime. Model routing is configured via models_catalog.yaml with policy-based rules matching by team, agent, and operation type, enabling mixed use of different LLM providers (OpenAI, Azure OpenAI, Ollama) without modifying agent code.

For deployment, the dev mode uses SQLite + ChromaDB with zero external dependencies, while production mode switches to PostgreSQL + pgvector + MinIO/S3, deployed via Helm Chart to Kubernetes. Local development supports Dev Container, authentication integrates Keycloak/OIDC, and workflow orchestration supports Temporal.

Core Capabilities#

• Agent Authoring & Runtime: ReAct/Profile Agents for focused tool-driven scenarios, Graph Agents for multi-step business workflows, both with streaming output and checkpoint support
• Knowledge & RAG: End-to-end document ingestion pipeline, ChromaDB for dev, pgvector for production, optional OpenSearch
• Model Routing: Named profiles + policy-based routing (matching by team_id, agent_id, user_id, operation, purpose), multi-provider mixing without code changes
• Security & Access: Keycloak/OIDC authentication + ReBAC (Relationship-Based Access Control) team permission model
• Dynamic Agent Creation: Runtime creation via code or API, mixing declarative YAML with programmatic flows

Architecture Overview#

Frontend (React, Vite, :5173)
        │
        ▼
Agentic Backend (FastAPI, :8000)  ◄── Agent runtime, session orchestration, model routing
        │↕
Knowledge Flow Backend (FastAPI, :8111)  ◄── Document ingestion, vectorization, RAG retrieval
        │↕
Control Plane Backend (FastAPI, :8222)  ◄── Team/user management, Agent registry, access policies

All backends are built on Python/FastAPI, with Alembic for database migrations and Pyright for type checking. Supports Temporal workflow engine, Kubernetes deployment via Helm Chart, and local K8s with k3d + Traefik Ingress.

Quick Start#

git clone https://github.com/ThalesGroup/fred.git
cd fred
cp agentic-backend/config/.env.template agentic-backend/config/.env
cp knowledge-flow-backend/config/.env.template knowledge-flow-backend/config/.env
# Edit .env with your API Key
make run-app  # Single-process mode
# Or: make run-multi  # Separated mode

Frontend: http://localhost:5173

Data Persistence#

Pending Confirmation#

• fred-deployment-factory public repo URL
• Specific Temporal workflow types
• MCP protocol version
• Observability implementation details (e.g., OpenTelemetry)
• Horizontal scaling capabilities of each backend component