This documentation is the product. There is no playground. There is no chat UI. If you want to evaluate HermesBridge, run the quickstart against a sandbox identity.

Documentation

HermesBridge is an OpenAI-compatible LLM gateway that requires cryptographically-attested agent identity for every request. The documentation is organized into five sections.

Getting started

Provision identity

Register an agent DID through HermesVault.

Install SDK

Python, TypeScript, Rust, and Go SDKs.

First signed request

Make your first authenticated LLM completion.

Migrating from OpenAI

Adapt existing code to the identity model.

Concepts

Identity model

DIDs, key binding, capability claims, attestation chains.

Attestation tiers

Self-attested, runtime-signed, TEE-verified.

Routing modes

Auto, cheapest, fastest, capability-based.

Caching behavior

Deterministic prompt caching and cache keys.

API reference

POST /v1/chat/completions

Primary completion endpoint.

POST /v1/embeddings

Text embeddings.

POST /v1/identity/attest

Submit or upgrade attestation evidence.

GET /v1/models

List available models and routing metadata.

GET /v1/usage

Per-agent usage records and audit log.

SDKs

Python

hermesbridge PyPI package.

TypeScript/Node

@hermesbridge/sdk npm package.

Rust

hermesbridge crates.io package.

Go

github.com/hermeslabs/hermesbridge-go.

Protocol

Request signing spec v0.1

Full protocol specification.

JWT claim structure

All supported claims and their semantics.

Error taxonomy

Complete error code reference.