forked from boos/BoosAPI
boos
5501558354
The previous approval registry used a oneshot channel, so two independent signals decided an approval: the HTTP handler's result (from Sender::send's bool) and the agent loop's allow/deny (from which timeout arm fired). At the 300s boundary these could disagree — a late /approve could see send() return Ok (receiver not yet dropped) and report ok:true to the user, while the timeout's final poll simultaneously fired and denied the tool. The frontend was told "approved" while nothing ran. The mirror case (ok:false but allowed) was impossible, so it never executed an unapproved tool, but the user-facing signal could lie. Replace the channel with a single source of truth: each entry is an AtomicU8 state machine (Pending -> Approved/Denied/TimedOut) plus a Notify. The decision is made by exactly one compare_exchange from Pending — whoever wins (approver or timeout) settles it, everyone else observes a losing CAS and no-ops. resolve() now returns whether *its* CAS won, so /approve tells the frontend the truth; the loop decides purely from the settled atomic. They read the same atomic and cannot disagree. A PendingGuard removes the registry entry on drop (normal completion or aborted/disconnected session), so there is no leak on any path. The callback also emits an approval_resolved SSE event with the final decision so a late click — already told ok:false — can dismiss its modal with the real outcome. Verified with an integration test that races resolve() against a near-zero timeout 5000x and asserts ok==true iff the executed decision is Approved (never a mixed state). Run via tests/ because the in-crate lib test binary is currently blocked by unrelated AppConfig test fixtures in other modules.
BoosAPI
AI Aggregation Gateway + Workflow Studio — one endpoint for many AI relay stations
Aggregate New API / One API / OneHub / DoneHub / Veloera / AnyRouter / Sub2API stations
into one API key and one compatible endpoint, with model discovery, smart routing, logs, billing, and failover.
Workflow Studio embeds the ComfyUI frontend and can run against Comfy Cloud, RunningHub, your own ComfyUI, or BoosAPI's API-native executor.
中文 | English | Deployment
Introduction
BoosAPI is a self-hosted Rust gateway for AI relay stations. It centralizes upstream sites, accounts, downstream API keys, model routes, usage logs, image/video generation, user dashboards, and Workflow Studio in one service.
Supported upstreams:
- Relay panels: New API, One API, OneHub, DoneHub, Veloera, AnyRouter, Sub2API
- Compatible APIs: OpenAI / Claude / Gemini compatible, cliproxyapi / CPA
- Preset providers: Aliyun, Zhipu, Doubao Coding Plan, DeepSeek, Moonshot, MiniMax, ModelScope
- OAuth connections: Codex, Claude, Gemini CLI, Antigravity
Key Features
Unified API Gateway
- OpenAI-compatible
/v1/chat/completions,/v1/responses,/v1/images/generations,/v1/models, and files endpoints - Claude-compatible
/v1/messages, including request/response and streaming event conversion - Smart token routing, channel cooldown, failover, model discovery, and usage accounting
- Admin and user dashboards with API key management and model visibility controls
Workflow Studio
Workflow Studio uses an embedded ComfyUI frontend. The execution backend is selected by comfyui.backend:
| backend | Purpose |
|---|---|
cloud |
Proxy Comfy Cloud API for cloud execution and subscription models |
runninghub |
Proxy RunningHub's ComfyUI-compatible API for its plugins, models, and workflows |
external |
Proxy a full ComfyUI instance you deploy yourself |
boosapi |
Use BoosAPI's Rust API-native workflow executor and configured model routes |
docker |
Local full ComfyUI mode for GPU machines and local model directories |
Cloud, RunningHub, external API, and BoosAPI executor modes do not require a GPU on the BoosAPI server. A GPU is only needed when you run local ComfyUI inference on the same machine.
Assistants
- Video / workflow assistant with Markdown chat, reasoning sections, attachments, and model route selection
- Picture assistant using
/v1/images/generations, background tasks, progress display, cancellation, and Workspace archive - Volcengine video task templates for create-task and query-task flows
Hardware Guide
| Mode | CPU | Memory | Disk | GPU |
|---|---|---|---|---|
Gateway, admin UI, /v1/* proxy only |
2 cores minimum, 4 recommended | 2GB minimum, 4GB recommended | 10GB+ | Not required |
| BoosAPI + Workflow UI + remote APIs / Comfy Cloud / RunningHub | 4 cores recommended | 4GB minimum, 8GB recommended | 20-50GB | Not required |
| External ComfyUI proxy | 4 cores recommended | 8GB recommended | 50GB+ | Not on BoosAPI; required by the ComfyUI machine as needed |
| Local Docker / local ComfyUI image generation | 4-8 cores | 16GB+ | 100GB+ | NVIDIA 8GB VRAM minimum, 12GB+ recommended |
| Local video / digital human / large workflows | 8 cores+ | 32GB minimum, 64GB recommended | 200GB-1TB | NVIDIA 24GB VRAM minimum, 48GB+ recommended |
Recommended SaaS gateway server:
4 CPU cores
8GB RAM
50GB SSD
No GPU required
Quick Start
cp config.toml.example config.toml
./build-and-run.sh
For a production bundle with embedded web assets:
./build-and-run.sh --production
The production output is intended to run with only:
boosapi
config.toml
Web assets and Workflow UI assets are embedded in the Rust binary when built for production.
Workflow Backend Configuration
Comfy Cloud:
[comfyui]
backend = "cloud"
base_url = "https://cloud.comfy.org"
api_key = "your-comfy-cloud-api-key"
RunningHub:
[comfyui]
backend = "runninghub"
base_url = "https://www.runninghub.cn/your-comfy-compatible-endpoint"
api_key = "your-runninghub-api-key"
External ComfyUI:
[comfyui]
backend = "external"
base_url = "http://127.0.0.1:8188"
BoosAPI API-native executor:
[comfyui]
backend = "boosapi"
Environment Variables
Configuration can be overridden with BOOSAPI_<SECTION>__<KEY>:
BOOSAPI_SERVER__HOST=0.0.0.0
BOOSAPI_SERVER__PORT=4000
BOOSAPI_AUTH__AUTH_TOKEN=your-admin-token
BOOSAPI_AUTH__PROXY_TOKEN=sk-your-proxy-token
BOOSAPI_AUTH__JWT_SECRET=your-jwt-secret
BOOSAPI_DB__URL=sqlite:data/boosapi.db?mode=rwc
Tech Stack
| Layer | Tech |
|---|---|
| Backend | Rust, Axum 0.8, Tokio |
| Database | SQLx, SQLite runtime path, PostgreSQL schema/migration preparation |
| Proxy engine | Token Router + Proxy Conductor |
| Protocols | OpenAI / Claude / Gemini compatible transformers |
| Frontend | React 19 + Vite |
| Workflow UI | Embedded ComfyUI frontend variants for cloud/external-compatible backends |
| Static serving | Rust binary serves dist/web and Workflow UI assets |
| Container | Docker / Docker Compose |
Node.js is only a build-time tool for the frontend. Production does not require a Node.js process.
Local Development
cargo check
cargo run
pnpm install
pnpm run dev:web
Build checks:
cargo check
pnpm run typecheck:web
npx vite build