Skip to content

POSITIONING.md

Latest commit

 

History

History
43 lines (29 loc) · 1.75 KB

POSITIONING.md

File metadata and controls

43 lines (29 loc) · 1.75 KB
Raw

Positioning

codex-flow is not trying to replace Codex, Aider, OpenHands, LangGraph, or a CI system.

It fills a narrower gap:

Codex App / CLI is good at doing agentic work, but a normal turn is linear. Maintainer tasks often need fan-out, replay, journals, and a clean way for Codex itself to generate and run temporary workflows.

What codex-flow is

  • A small workflow layer for Codex App / CLI.
  • A natural-language trigger via the bundled Codex skill.
  • A runtime for parallel, pipeline, phase, and structured agent calls.
  • A replayable journal for interrupted or repeated work.
  • A maintainer automation tool for bug investigation, PR review, issue triage, and release checks.

What codex-flow is not

  • Not a general graph framework like LangGraph.
  • Not a full autonomous developer environment like OpenHands.
  • Not a code-editing chat interface replacement like Aider.
  • Not a CI runner or task queue.
  • Not a model provider abstraction layer.

Why use it

Use codex-flow when the task has independent branches or repeated stages:

  • review these N files independently,
  • investigate these hypotheses in parallel,
  • run the same analysis across several issues,
  • generate then review per-item patches,
  • resume a long Codex-backed workflow without paying for completed nodes again.

Do not use it for one small edit or one simple question. A direct Codex turn is better there.

Design bet

The core bet is that the workflow should be generated by Codex at the edge, but executed by a deterministic engine:

  • Codex understands the user goal and writes the workflow.
  • The engine handles concurrency, replay, schema validation, budgets, journaling, and backend quirks.
  • The maintainer gets a resumable result instead of a long unstructured chat transcript.