Fred
Learn

Learn

What Fred Is

Fred is a fully open-source platform for deploying intelligent, tool-using AI assistants in production environments. It is designed for organizations that operate on Kubernetes and cloud-native infrastructure, where governance, observability, and operational discipline are as critical as model quality.

Fred is not a prototyping framework. It is a complete, integrated system — from the user interface to the document knowledge pipeline — built to run in real deployments, under real operational constraints.

The Problem Fred Addresses

The landscape of agentic AI frameworks is expanding rapidly. Most available solutions — whether commercial or open source — excel at one dimension: developer experience, visual workflow design, or ecosystem breadth. Few are designed from the ground up for production operations in regulated, security-sensitive, or large-scale enterprise contexts.

Organizations deploying AI assistants at scale face a consistent set of challenges:

  • Model governance — controlling which model is used, under which conditions, and by whom
  • Knowledge management — ingesting, processing, and maintaining large, heterogeneous document corpora with full traceability
  • Operational transparency — understanding what agents do, why they do it, and where failures occur
  • Security and access control — enforcing boundaries at the platform level, not the application level

Fred was built to address these challenges as a platform, not as a collection of integrations.

Architecture Overview

Fred is composed of three independently deployable components, each with a well-defined API surface.

Agentic Backend orchestrates AI agents, manages user sessions, routes requests to the appropriate model, and handles tool invocation. Agent logic is expressed as explicit, testable state machines using LangGraph. Tools — whether built-in, MCP-based, or custom — are typed and auditable.

Knowledge Flow Backend manages the complete document lifecycle: ingestion, format conversion, OCR, vectorization, metadata enrichment, and semantic retrieval. It supports heterogeneous document corpora at scale and is designed to be extended with custom processing pipelines.

React Frontend provides a production-ready interface for end users and administrators: session history, inline source citations, document libraries, and configuration panels. It is a functional starting point, not a demo.

All three components run as containerized services on Kubernetes. Temporal handles durable, fault-tolerant workflow execution. Standard observability integrations (metrics, logs, traces) are included by design.

Design Principles

Fred is built around a small number of consistent principles that distinguish it from lighter-weight alternatives.

Tool-first agents. Agents interact with the world through typed, auditable tool calls. Behavior is explicit and inspectable, not implicit in model outputs.

Catalog-driven configuration. Models, agents, and routing rules are declared in YAML catalogs. Runtime behavior is separated from application code, which makes configuration changes safe and reviewable.

Governance as a platform capability. Access control, model selection policies, and data scope boundaries are enforced at the platform level. Individual application teams do not need to re-implement these controls.

Cloud-native by design. Fred is built for Kubernetes from the start — not adapted to it. Deployment, scaling, and operational patterns follow cloud-native conventions.

Fully open source. Fred is released under Apache 2.0. There are no usage tiers, feature gates, or proprietary connectors. What you see is what you get.

Who Should Consider Fred

Development and platform teams seeking a structured, extensible framework for building and operating AI agents in Python, on Kubernetes, without vendor lock-in.

Infrastructure and security architects evaluating open-source agentic infrastructure where governance, auditability, and operational maturity are primary selection criteria.

Product and business stakeholders who need a complete foundation that supports both early experimentation and long-term production deployment — with a clear path from prototype to scale.

Organizations in regulated industries where data sovereignty, access control, and audit trails are non-negotiable requirements.

Go further