Open source LLM gateways, installed and run
LLM and AI gateways you can run yourself. The list is ordered by whether each project came up when Argusic installed it fresh, with the recording of that attempt one click away.
Tested between and . Each row shows its own test date; a project can change after that day.
14 of 19 tested projects run. 27 more waiting for a test.
In short: 11 of the 19 tested projects started as-is on a fresh machine: 9router, casdoor, gateway, coai, plano, gpt-load, CPA-Manager-Plus, and krakend-ce, and 3 more. 3 more started once a stand-in replaced a service they expect, such as a database: ccs, future-agi, and APIPark. 5 could not be verified: higress, router, orbit, agentgateway, and agent-router; the log shows where each one stopped.
Measured by Argusic on a fresh machine every time. Every number links to its evidence.
| # | project | verdict | Argusic Score | language | stars | tested on |
|---|---|---|---|---|---|---|
| 1 | 9router | Runs | 100 / 100 | JavaScript | 30,455 | |
Unlimited FREE AI coding. Connect Claude Code, Codex, Cursor, Cline, Copilot, Antigravity to FREE Claude/GPT/Gemini via 40+ providers. Auto-fallback, RTK -40%... What the test found: Root npm install completed, npm run build succeeded, the standalone Next.js app serves the login page (200), models API (200 with JSON model list), and version endpoint (200), and the test suite runs 2562 passing tests. 42 minutes. | ||||||
| 2 | casdoor | Runs | 100 / 100 | Go | 14,524 | |
An open-source Agent-first Identity and Access Management (IAM) /LLM MCP & agent gateway and auth server with web UI supporting OpenClaw, MCP, OAuth, OIDC... What the test found: Casdoor server runs on port 8000 with SQLite, frontend serves index.html (200), login authenticates admin/123 via JSON API, OpenID Connect configuration endpoint returns valid metadata, and the API returns organization and user data from the demo seed. 14 minutes. | ||||||
| 3 | gateway | Runs | 100 / 100 | TypeScript | 13,148 | |
A blazing fast AI Gateway with integrated guardrails. Route to 1,600+ LLMs, 50+ AI Guardrails with 1 fast & friendly API. What the test found: npm install and build succeed, the Node.js server starts on port 8787 and responds 200 on /v1, 170 unit tests pass. 30 minutes. | ||||||
| 4 | coai | Runs | 100 / 100 | TypeScript | 9,322 | |
🚀 Next Gen Multi-tenant AI One-Stop Solution. Builtin Admin & Billing System. Enterprise-Grade Unified LLM Gateway Support for 200+ Models And 35+ Providers... What the test found: CoAI.Dev builds and runs on port 8094 using SQLite (no MySQL) and a fake Redis mock, with login, user info, admin analytics, quota, subscription, channel management, and model listing endpoints all responding correctly. 18 minutes. | ||||||
| 5 | plano | Runs | 100 / 100 | Rust | 7,082 | |
Plano is an AI-native proxy server and data plane for agentic apps. Smart LLM routing, observability, agent orchestration, and guardrails so you stay focused... What the test found: Plano native mode starts, health endpoint returns 200, CLI reports version 0.4.36, all Rust and Python tests pass. 21 minutes. | ||||||
| 6 | gpt-load | Runs | 100 / 100 | Go | 7,055 | |
Self-hosted AI gateway for multi-channel, multi-credential setups, API keys and subscription accounts, scheduling, failover, request logs and usage. 自托管 AI... What the test found: GPT-Load 2.0 binary builds, web UI compiles, server starts on 127.0.0.1:3001, health endpoint returns HTTP 200, auto-generates AUTH_KEY, and shuts down gracefully with all database migrations completed and catalog synced. 8 minutes. | ||||||
| 7 | CPA-Manager-Plus | Runs | 100 / 100 | Go | 3,761 | |
A self-hosted CPA / CLIProxyAPI management panel and AI gateway observability dashboard for requests, usage, cost, quota, failures, and account health. What the test found: CPA Manager Plus web app installs, builds, type-checks, and runs all 4156 web tests passing; repo-level tests pass 256/257. The dev server serves the management panel on port 5173 and returns HTTP 200. 12 minutes. | ||||||
| 8 | krakend-ce | Runs | 100 / 100 | Go | 2,689 | |
KrakenD Community Edition: High-performance, stateless, declarative, API Gateway written in Go. What the test found: KrakenD v2.13.11 binary built, all 58 integration tests pass, server listens on port 8080 and responds to /__health with 200. 4 minutes. | ||||||
| 9 | OrcaRouter-Lite | Runs | 100 / 100 | Python | 1,729 | |
Self-hosted LLM router with a managed safety net. OpenAI-compatible. BYOK. Single-workspace. Streaming. For more advanced routing choose hosted OrcaRouter What the test found: OrcaRouter Lite installs from source, all 703 tests pass, the server starts on port 8000, seeds a workspace+key, and serves /health, /v1/models with auth gating. 7 minutes. | ||||||
| 10 | EnterpriseAgentFramework | Runs | 100 / 100 | Java | 845 | |
ReachAI企业级智能体开发平台：快速、安全完成已有业务系统智能化改造，让 AI 在 OA、ERP、CRM 等原系统中查数据、填表单、办业务。ReachAI: Quickly and securely bring AI to existing enterprise systems, enabling AI to... What the test found: Backend builds cleanly and all 2246 unit/integration tests pass across 9 Maven modules. Frontend builds with npm run build and 453 of 454 vitest tests pass across 10 test suites (1 pre-existing timezone-dependent dashboard test fails in UTC). 34 minutes. | ||||||
| 11 | paritok-4b-v1 | Runs | 96 / 100 | Python | 1,453 | |
Non-destructive compression gateway for AI coding agents. Cuts token bills 25% on turn 1 to past 85% in long or saturated sessions, and fits ~3× more turns in... What the test found: Python package installed in a venv, the proxy serves on port 8088, answers /health (200) and /stats (200) and /v1/messages/count_tokens (200), and all 316 tests in the test suite pass. 8 minutes. | ||||||
| 12 | ccs | Runs with mocks | 92 / 100 | TypeScript | 2,876 | |
Switch between Claude accounts, Gemini, Copilot, OpenRouter (300+ models) via CLIProxyAPI OAuth proxy. Visual dashboard, remote proxy support, WebSearch... What the test found: The project builds, the ccs CLI launches without config errors, and the tool-sanitization proxy plus related integration tests (75 tests across 5 files) pass cleanly. 18 minutes. | ||||||
| 13 | future-agi | Runs with mocks | 92 / 100 | Python | 2,124 | |
Open-source, end-to-end platform for evaluating, observing, and improving LLM and AI agent applications. Tracing · Evals · Simulations · Datasets · Gateway ·... What the test found: Frontend dependencies installed with yarn, unit and integration tests pass (11/11) using jsdom environment. Backend dependencies installed via uv sync (441 packages). Django 5.1.8 and Django REST framework import correctly. Backend cannot fully start or run tests without Docker for PostgreSQL, ClickHouse, Redis, and... 50 minutes. | ||||||
| 14 | APIPark | Runs with mocks | 92 / 100 | TypeScript | 1,820 | |
Cloud native, ultra-high performance AI&API gateway, LLM API management, distribution system, open platform, supporting all AI APIs.🦄云原生、超高性能 AI&API网关，LLM API... What the test found: APIPark binary builds, starts on port 8288 with MySQL and Redis, serves HTTP 200 JSON API responses and embedded index.html, auto-creates all database tables, and passes 4 test packages. 57 minutes. | ||||||
| 15 | higress | Could not verify | 40 / 100 | Go | 9,498 | |
🤖 AI Gateway | AI Native API Gateway What the test found: Go controller binary and hgctl CLI build and respond to help; all Go unit tests pass (26 packages); WASM AssemblyScript SDK builds both debug and release wasm binaries. 15 minutes. | ||||||
| 16 | router | Could not verify | 40 / 100 | Go | 5,580 | |
Model router for agentic systems. Routes every prompt to the right model in <50ms. Cut costs 40-70% with just an endpoint change. What the test found: could not be verified; the log shows where it stopped. 14 minutes. | ||||||
| 17 | orbit | Could not verify | 10 / 100 | Python | 353 | |
Self-hosted AI gateway for private RAG, natural-language data access, and tool-calling agents. What the test found: could not be verified; the log shows where it stopped. 28 minutes. | ||||||
| 18 | agentgateway | Could not verify | 0 / 100 | Rust | 5,228 | |
Next Generation Agentic Proxy for AI Agents and MCP servers What the test found: could not be verified; the log shows where it stopped. 87 minutes. | ||||||
| 19 | agent-router | Could not verify | 0 / 100 | Go | 2,199 | |
Manages Unified Access to Generative AI Services built on Envoy Gateway What the test found: could not be verified; the log shows where it stopped. 87 minutes. | ||||||
| - | TokenHub | Not yet tested | - | Go | 1,353 | - |
| - | litellm-agent-control-plane | Not yet tested | - | Rust | 1,305 | - |
| - | GoModel | Not yet tested | - | Go | 1,226 | - |
| - | copilot-api | Not yet tested | - | TypeScript | 1,075 | - |
| - | ThinkWatch-Lite | Not yet tested | - | TypeScript | 1,070 | - |
| - | CliRelay | Not yet tested | - | Go | 1,030 | - |
| - | temps | Not yet tested | - | Rust | 826 | - |
| - | casbin-gateway | Not yet tested | - | Go | 642 | - |
| - | Lynkr | Not yet tested | - | JavaScript | 621 | - |
| - | dario | Not yet tested | - | JavaScript | 560 | - |
| - | otari | Not yet tested | - | Python | 513 | - |
| - | gemini-web2api-go | Not yet tested | - | Go | 485 | - |
| - | sagify | Not yet tested | - | Python | 442 | - |
| - | kiji-proxy | Not yet tested | - | Go | 436 | - |
| - | ccLoad | Not yet tested | - | Go | 418 | - |
| - | privacy-filter | Not yet tested | - | Go | 339 | - |
| - | grok2api-egress-enhancements | Not yet tested | - | Go | 320 | - |
| - | llmio | Not yet tested | - | Go | 317 | - |
| - | digiRunner-Open-Source | Not yet tested | - | Java | 307 | - |
| - | gptGrok2api | Not yet tested | - | Python | 276 | - |
| - | 1flowbase | Not yet tested | - | Rust | 271 | - |
| - | ai-gateway | Not yet tested | - | Go | 269 | - |
| - | bella-openapi | Not yet tested | - | Java | 238 | - |
| - | bitrouter | Not yet tested | - | Rust | 235 | - |
| - | DoulorCloud | Not yet tested | - | TypeScript | 222 | - |
| - | yolorouter | Not yet tested | - | Go | 215 | - |
| - | proxy | Not yet tested | - | TypeScript | 205 | - |
runs installed and started with its real dependencies. runs with mocks started after stand-ins replaced external services such as a database or a third-party API. could not verify neither the standard agent nor the stronger one got it running within the time limit; the log shows where it stopped.
How we tested
On this list as of the latest test: 11 projects ran as-is, 3 with mocks, 5 could not be verified, 27 still waiting. Languages tested: Go, Java, JavaScript, Python, Rust, TypeScript. Every attempt used a clean single-use machine, the subject at a pinned version, and a 45-minute limit; the complete procedure is on the methodology page.
Frequently asked questions (FAQs)
How is this list ranked?
By measurement, not opinion: projects Argusic installed and launched on a fresh machine come first, then those that ran with mocks in place of external services, then those it could not verify. Ties go to the Argusic Score, then how popular it is on its own source.
Why are some projects unranked?
27 projects are still waiting for a test or for a finished attempt. They are listed without a rank until Argusic has measured them.
Where is the evidence?
Every row links to the project's Argusic page, where each run has a full log and a terminal recording stored with a sha256 fingerprint. The same pages exist for every one of the tested projects, on this list or not.
More lists in this category
- API gateways (shares APIPark, CliRelay, agent-router, agentgateway, bitrouter, ccLoad, digiRunner-Open-Source, gpt-load, higress, krakend-ce with this list)
- self-hosted AI apps (shares CliRelay, Lynkr, ccLoad, gpt-load, orbit, otari with this list)
- open source coding agents (shares EnterpriseAgentFramework, casbin-gateway, paritok-4b-v1, router with this list)
- observability tools (shares CPA-Manager-Plus, GoModel, future-agi with this list)
- AI agent frameworks (shares 9router, EnterpriseAgentFramework with this list)
- CI/CD tools (shares temps with this list)
- MCP servers (shares agentgateway with this list)
- speech tools (shares orbit with this list)
- vector databases (shares orbit with this list)
- API clients
- CMS platforms
- VPN tools
- backup tools
- browser automation tools
- code editors
- Open source databases, installed and queried
- developer CLI tools
- e-commerce platforms
- ebook readers
- game engines
- open source games
- home automation tools
- low-code platforms
- map tools
- message queues
- Open source music servers, installed and played
- Open source office suites, installed and launched
- self-hosted password managers
- project management tools
- screen recorders
- search engines
- self-hosted analytics
- static site generators
- uptime monitors
- open source video editors
- video players
- web scraping tools
- whiteboard tools
- wikis
- workflow automation tools
- self-hosted Notion alternatives
- self-hosted dashboards
- self-hosted git servers