Open-source AI gateway
A unified layer to control providers, routes, and operational continuity.
ForgeRouter centralizes AI requests, monitors upstream health, selects best endpoints, and applies automatic fallback. Client apps get a rock-solid interface while operations retains total routing governance.
- Problem
- Dependency and fragmentation across AI providers
- Evidence
- Live system, screens, documentation, and source code published
- Usage
- Autonomous evaluation or guided enterprise deployment
Python + FastAPI, PostgreSQL with YAML fallback. Deploys via Docker Compose on your own server — zero third-party cloud lock-in.
The problem
When every application connects directly to every provider, switching models costs far too much.
API keys, schemas, models, and retry logic scatter across your stack. A single provider outage becomes an incident across all apps at once.
Tightly coupled integrations
Every internal app needs provider-specific quirks and redundant error handling.
Fragmented consumption
Auditing which application consumed which model and via which route is difficult.
Single point of failure
When an upstream provider drops, consuming software halts or requires rushed workarounds.
What ForgeRouter solves
A single gateway shielding all AI providers you consume — local, API key, or OAuth/subscription based.
Multiple providers, single gateway
Local Ollama plus API key providers (Groq, OpenRouter, Mistral, NVIDIA, Cloudflare, Cohere, Gemini Studio, GitHub Models) or subscription logins (Claude Code, Codex, Antigravity, DeepSeek, Z.ai) — unified in a single pool.
Health checks & auto fallback
Requests evaluate candidate endpoints sequentially — failing, rate-limited, or degraded providers are skipped in favor of healthy targets without breaking client requests.
Smart demand routing
Classifies incoming prompts by intent (simple, complex, reasoning, vision, audio, coding) to route to optimal models — preserving expensive quota.
3 protocols, 1 pipeline
Native support for OpenAI Chat Completions, Anthropic Messages (Claude Code), and OpenAI Responses (Codex CLI) — true incremental streaming throughout.
Embedded admin dashboard
Built-in React GUI to manage providers, agents, routes, and metrics — served directly by the API with no extra container.
Per-agent keys and budgets
Each consumer application receives dedicated API keys, model restrictions, and monthly cost ceilings — without storing prompt content.
How to deploy
Standalone automated installer with embedded PostgreSQL — zero manual setup required:
Why it is open source and free
Tools that solve real challenges should not hide behind paywalls. ForgeRouter is the exact software Darckware uses daily to route AI workloads between local, free, and commercial models. Releasing it openly is how we demonstrate engineering quality in practice. MPL-2.0 license, free to use, zero licensing costs, deployed entirely on your own infrastructure.
Documentation
Workflow, setup guide, verification steps, and production screenshots.
Complete documentation walks through routing mechanics, installation procedures, and screenshots of all four administrative screens.
Saw quality in our engineering and want to build something?
Lara understands what caught your attention, identifies your expected outcome, and structures context for a technical conversation.