Build agents that actually work.
A platform that brings development and business teams together. Iterate fast. Reach production. Improve continuously.
Hebo Cloud | Docs | Blog | Discord
- Gateway — Unified access to state-of-the-art LLMs with native support for
/chat/completions(OpenAI),/responses(OpenResponses), and/messages(Anthropic). Embeddings and multi-turn conversations out of the box. - Observability — Conversation-level traces with token and latency tracking, tool call inspection, and full conversation replay — not just individual request logs.
- Evaluations — Define evals as simple Markdown files. Test agent behavior against guidelines, tool usage, and hand-off triggers.
Anthropic, OpenAI, Amazon Bedrock, Google Vertex, Microsoft Azure, DeepSeek, xAI, Alibaba (Qwen), MiniMax, Z.ai (Zhipu), Moonshot, Fireworks, DeepInfra, Together AI, Chutes, Groq, and Voyage.
All providers support BYOK (Bring Your Own Key). Platform-level keys are optional — users can configure their own credentials per organization.
Browse the full model catalog: gateway.hebo.ai/v1/models
Sign up at console.hebo.ai and start sending requests through the gateway — no infrastructure needed.
import OpenAI from "openai";
const hebo = new OpenAI({
baseURL: "https://gateway.hebo.ai/v1",
apiKey: process.env.HEBO_API_KEY,
});
const chat = await hebo.chat.completions.create({
model: "claude-sonnet-4-5",
messages: [
{ role: "system", content: "You are a helpful assistant." },
{ role: "user", content: "What is the Hebo gateway?" },
],
});Run everything locally with Docker Compose — a single container with all services included.
cd infra/self-hosted
docker compose up -dOpen http://localhost:8520 to access the console.
Create a .env file in infra/self-hosted/ and set the variables you want to override. At minimum for production:
| Variable | Why |
|---|---|
AUTH_SECRET |
Replace the default with a strong random string |
OPENAI_API_KEY, ANTHROPIC_API_KEY, … |
Platform-wide LLM provider keys (users can also bring their own) |
Optional: OAuth (GITHUB_CLIENT_ID, GOOGLE_CLIENT_ID, …) and SMTP (SMTP_HOST, SMTP_PORT, …) for passwordless login and invitations.
See infra/self-hosted/.env.example for the full variable reference.
| Hebo | Langfuse | Helicone | Portkey | LiteLLM | OpenRouter | Vercel AI | |
|---|---|---|---|---|---|---|---|
| Gateway | Stateful* | — | Stateless | Stateless | Stateless | Stateless | Stateless |
| Observability | Conversation view | Request level | Request level | Request level | 3rd party | 3rd party | 3rd party |
| Evaluations | Simple Markdown | Yes | Via integrations | Yes | — | — | — |
| User experience | Dev & Business | Dev-focused | Dev-focused | Dev-focused | Dev-focused | API only | API only |
| Deployment | Cloud & self-host | Cloud & self-host | Cloud & self-host | Cloud only | Self-host | Cloud only | Cloud only |
* Stateful: Hebo supports /responses and /conversations endpoints, enabling multi-turn sessions and persistent conversation history without managing state in your application.
- Discord — ask questions, share feedback
- X (@heboai) — follow for updates
- Docs — guides and API reference
- Blog — product updates, guides, and research notes
We welcome contributions! See CONTRIBUTING.md for development setup, architecture, and guidelines.