thinkany-ai / autojev

AutoJev 
 English · 简体中文 
 One local gateway for your AI agents. Route every request to the right model. 
 Download ·
 Website ·
 Issues ·
 MIT 
AutoJev is a desktop app built with Tauri, React, and Rust. Manage providers, models, and routing rules in one place, and connect agents such as Codex, Claude Code, and Hermes through a single local endpoint.
> AutoJev is in early development. Available installers and supported release platforms are listed in GitHub Releases.
## Features
- **Providers and models**: Manage API endpoints, keys, pricing, image input support, and context lengths. Image and context settings are informational and do not restrict requests.
- **Model routing**: Choose models with intelligent selection or load balancing, with speed, cost, quality, or balanced preferences. Includes session affinity, failover, and cooldowns.
- **Decision model**: Use OpenRouter Jev to select a candidate model, with local routing as a fallback when Jev is unavailable.
- **Protocol conversion**: Support for OpenAI Responses, Chat Completions, and Anthropic Messages, including streaming and function tool calls. See protocol compatibility for limitations.
- **Agent integration**: Detect local agents, remember model and route selections, back up and restore configurations, and reconnect agents on startup according to saved preferences. Model picker support varies by client.
- **Debugging and statistics**: A debug console, model speed tests, request logs, token usage, and estimated costs.
- **Desktop updates**: Automatic update checks, user-initiated downloads and installation, signature verification, and restart to update in release builds.
 [code] 
## Inside AutoJev
Add providers, measure models, build routes, connect agents, then inspect requests and usage.
### Manage providers in one place
Add OpenRouter, DeepSeek, or compatible API endpoints, or import existing providers from CC Switch and Termany. Test, enable, or disable each provider while keeping its configuration saved locally.
### Measure models before routing
Compare first-response latency and output speed with built-in speed tests. Configure input and output pricing, and keep image support and context-window information alongside each model.
### Create reusable routes
Group candidate models into routes such as `fast`, `quality`, or `cheap`, exposed as `autojev/ `. Choose intelligent selection or load balancing, with session affinity, failover, and cooldowns.
### Connect your agents
See detected agents, configuration paths, and connection status in one view. Select models or routes per agent, batch-update connected agents, and restore backed-up configurations when disconnecting.
### Test requests in the debug console
Try models and routes in conversation or cURL mode. Switch API protocols and inspect the selected model, timing, token usage, estimated cost, and raw JSON response.
### Inspect logs and usage
Filter request logs by time, agent, provider, and status to investigate routing and fail