OpenAI Integration¶

Codex CLI¶

The CODEX.md file at the repo root gives the OpenAI Codex CLI full project context: module map, build commands, invariants, and what not to change. Clone the repo and Codex is immediately project-aware.

Assistants API / Responses API¶

tools/openai-tools.json contains OpenAI function-calling schemas for diff_traces, inspect_trace, and record_snippet. Paste directly into your assistant definition:

import json
import openai

tools = json.loads(open("tools/openai-tools.json").read())

response = openai.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Did my agent regress between runs?"}],
    tools=tools,
)

GPT Actions / Custom GPTs¶

The openapi.yaml at repo root is a complete OpenAPI 3.1 spec for a REST wrapper.

Run the server locally¶

pip install "agentdelta[api]"
uvicorn agentdelta.api:app --reload
# Server at http://localhost:8000
# Docs at  http://localhost:8000/docs

Register as a ChatGPT Action¶

Deploy the server to a public URL (Railway, Fly.io, Render, etc.)
In ChatGPT → My GPTs → Create → Configure → Add actions
Import from URL: https://your-deployment.example.com/openapi.yaml
Set authentication if needed
Save and test

API endpoints¶

Method	Path	Description
`POST`	`/diff`	Compare two traces, returns DiffResult JSON
`POST`	`/inspect`	Summarise a single trace
`GET`	`/health`	Liveness probe — returns version

Interactive docs are available at /docs (Swagger UI) and /redoc when the server is running.