AnthMorph

AnthMorph is a high-performance Rust API bridge for modern AI coding clients. Its main job is to let Codex and codex-vl use providers that do not expose the latest OpenAI /responses API by translating, normalizing, and streaming across Anthropic Messages, OpenAI legacy chat-completions, and Responses-style traffic.

Use it when a provider is powerful but speaks the wrong wire format for your client. AnthMorph keeps the local client surface stable, adapts model names and tool calls, applies local auth/rate-limit policy, and forwards to the selected backend with minimal runtime overhead.

Project Status

• Current line: 0.2.4
• Primary Codex API: POST /v1/responses
• Anthropic ingress: POST /v1/messages, POST /v1/messages/count_tokens
• OpenAI legacy ingress: POST /v1/chat/completions, POST /chat/completions
• Operational APIs: GET /v1/models, GET /health
• Target clients: Codex and codex-vl latest, plus Anthropic/OpenAI-compatible clients
• Target platforms: macOS, Linux, and Termux

What It Does

• Translates Codex/codex-vl Responses traffic to native Responses or legacy chat-completions backends.
• Accepts Anthropic Messages API traffic and adapts it to OpenAI-compatible or Anthropic-compatible backends.
• Accepts OpenAI legacy chat-completions traffic for clients and tools that still use /chat/completions.
• Streams Server-Sent Events for supported streaming paths.
• Normalizes configured/default models for clients that send default or Claude-style model names.
• Shortens long function/tool names for providers with stricter tool-name limits.
• Supports local ingress auth, backend API key forwarding, CORS allow-lists, and per-client rate limiting.
• Serves /v1/models from backend discovery plus a local fallback model cache.

Install

Global npm install:

npm install -g @mmmbuto/anthmorph

On macOS, npm install builds the local binary with Cargo.

Local source build:

cargo build --release

Run against a legacy OpenAI-compatible chat backend:

ANTHMORPH_BACKEND_URL=https://api.example.com/v1 \
ANTHMORPH_BACKEND_PROFILE=openai-generic \
ANTHMORPH_UPSTREAM_API=chat-completions \
ANTHMORPH_PRIMARY_MODEL=example/model \
ANTHMORPH_API_KEY="$PROVIDER_API_KEY" \
PORT=9876 \
anthmorph

Run against a native Responses backend:

ANTHMORPH_BACKEND_URL=https://api.example.com/v1 \
ANTHMORPH_BACKEND_PROFILE=openai-generic \
ANTHMORPH_UPSTREAM_API=responses \
ANTHMORPH_PRIMARY_MODEL=example/model \
ANTHMORPH_API_KEY="$PROVIDER_API_KEY" \
PORT=9876 \
anthmorph

Codex and codex-vl

Point Codex or codex-vl at AnthMorph with Responses wire format:

model = "example/model"
model_provider = "anthmorph"

[model_providers.anthmorph]
name = "Provider via AnthMorph"
base_url = "http://127.0.0.1:9876/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"
requires_openai_auth = false

Codex sends /v1/responses to AnthMorph. AnthMorph then chooses the configured upstream mode:

ANTHMORPH_UPSTREAM_API=responses          -> {BACKEND_URL}/responses
ANTHMORPH_UPSTREAM_API=chat-completions   -> {BACKEND_URL}/chat/completions

Runtime Surface

POST http://127.0.0.1:9876/v1/responses
POST http://127.0.0.1:9876/v1/messages
POST http://127.0.0.1:9876/v1/messages/count_tokens
POST http://127.0.0.1:9876/v1/chat/completions
POST http://127.0.0.1:9876/chat/completions
GET  http://127.0.0.1:9876/v1/models
GET  http://127.0.0.1:9876/health

Environment

Minimum direct environment:

PORT=9876
ANTHMORPH_BACKEND_URL=https://api.example.com/v1
ANTHMORPH_BACKEND_PROFILE=openai-generic
ANTHMORPH_UPSTREAM_API=chat-completions
ANTHMORPH_PRIMARY_MODEL=example/model
ANTHMORPH_API_KEY=...

Optional:

ANTHMORPH_REASONING_MODEL=...
ANTHMORPH_INGRESS_API_KEY=...
ANTHMORPH_ALLOWED_ORIGINS=https://example.test
ANTHMORPH_RATE_LIMIT_PER_MINUTE=60
ANTHMORPH_STRICT_MODEL=true
ANTHMORPH_STREAM_CHUNK_TIMEOUT_SECS=30

Validation

Local Rust tests:

cargo test

Docker release checks:

./scripts/docker_release_checks.sh

Docs

• Packaging details: docs/PACKAGING.md
• Release guide: docs/RELEASE.md
• Changelog: CHANGELOG.md

License

Apache-2.0. See LICENSE.

Copyright (c) 2026 DioNanos

Contact

Maintained by DioNanos.

• General / dev: [email protected]
• Security disclosures: [email protected]
• Project hub: https://mmmbuto.com