RelayForge

01 / Context

Why It Exists

In daily AI workflows, switching between ChatGPT, Claude, local models, and other services is frequent. Each service has different API formats, authentication methods, and rate limits. RelayForge provides a unified entry point, making model selection flexible and controllable.

Problem

Multiple AI models have inconsistent APIs, making switching costly. Some models have rate limits or regional restrictions, causing direct calls to fail. Sensitive information in requests (API keys, personal data) needs redaction before safe forwarding.

Solution

RelayForge runs a unified proxy gateway locally, wrapping different model APIs into a single interface. Supports fallback routing (auto-switch on primary failure), combo routing (distribute tasks across models), request redaction, and status diagnostics. Binds to localhost by default, no telemetry, prompts not logged, credentials redacted in UI and logs.

02 / System

Core Capabilities

• Unified multi-model API as single entry point, compatible with OpenAI and Anthropic interfaces
• Supports Ollama/LM Studio local models and cloud providers
• Combo routing, fallback, round_robin, weighted_round_robin
• Request privacy: redact or replace sensitive information before forwarding
• Lightweight usage analytics: local statistics on call counts and token usage
• Overview, Providers, Combo Models, Clients, Usage, Diagnostics, Settings Dashboard

03 / Workflow

Workflow

1. Configure model providers and routing rules
2. Application sends requests via unified API (default 127.0.0.1:18765/v1)
3. RelayForge selects model based on routing rules
4. Automatic fallback or combo distribution
5. Return results and record usage statistics

04 / Boundaries

Security & Privacy

RelayForge runs locally, binding to localhost by default. No telemetry, prompts not logged, credentials redacted in UI and logs. Only recent request metadata is stored. Uses user-configured provider credentials; no subscription session passthrough.

Current Limitations

• Requires user to configure providers and routing
• Gateway cannot bypass upstream quotas or regional restrictions
• Privacy boundaries depend on the ultimately chosen upstream model

05 / Roadmap

Roadmap

06 / Participate

How to Participate

Available on GitHub via Windows zip, cross-platform Node, or Docker/GHCR. Issues and pull requests are welcome. If you have specific multi-model collaboration needs, describe your use case in an issue.