adaptorch

Your AI says it works. AdaptOrch checks.

AdaptOrch runs AI-written patches in a clean project copy, separates code failures from runner failures, and returns one replayable receipt. Multi-model runs get the same receipt: which model answered, where they disagreed, and what it cost.

This distribution is a meta-package. It contains no code of its own; it installs the two thin clients for the hosted service at https://adaptorch.com:

pip install adaptorch          # or: uv tool install adaptorch
export ADAPTORCH_API_KEY="ado_..."   # create one at https://adaptorch.com/app/api-keys
adaptorchctl whoami

import os

from adaptorch_client import AdaptOrchClient, ClientConfig

client = AdaptOrchClient(
    ClientConfig(api_url="https://adaptorch.com", api_key=os.environ["ADAPTORCH_API_KEY"])
)
print(client.capabilities())

Using a coding agent instead of Python? Nothing to install:

claude mcp add --transport http adaptorch https://adaptorch.com/mcp \
  --header "Authorization: Bearer ${ADAPTORCH_API_KEY}"

Cursor, Codex, Gemini CLI, VS Code and Windsurf snippets: https://adaptorch.com/mcp-docs.

What is not in here

The verification engine and the control plane are not distributed as a package; they run in the hosted service. Releases 0.1.1 and 0.1.2 of this name were earlier reference implementations and are withdrawn.

License

Proprietary — Copyright (c) 2026 ClassicMate. All rights reserved. See LICENSE.