How it works

Start every Codex session inside a disciplined repo.

CodexFlow installs a versioned workflow layer your repo keeps — memory, skills, commands, runbooks, guardrails — through a signed paid setup.

  • One command
  • No SaaS lock-in
  • Git-reviewable
  • Signed paid payload
Before · After

The same task — without and with CodexFlow.

Vanilla Codex sessions start cold every time. CodexFlow sessions start from project memory, AGENTS.md rules, and reusable workflow commands.

Before · vanilla Codex

Re-explain the project, every session.

You paste rules, paths, and conventions into the chat. The next session won't remember.

$ codex> We're on Next.js 15, app router, Supabase, Stripe.
   Don't touch .env*, migrations/, or the webhooks.
   Use pnpm. Tests live in /tests. Run `pnpm build`
   before claiming done. Now please add SEO checks
   before shipping.…20 minutes later, after the model wanders…⚠ Modified .env.local · scaffolded /scripts/seo.ts  next session: starts again from zero
After · with CodexFlow

One short prompt. The repo carries the rules.

Codex reads AGENTS.md, picks the right skill and runbook, and respects the guardrails.

$ codex> /plan-feature add SEO checks before shipping→ loaded AGENTS.md · .codex/skills/nextjs-seo-audit→ guardrails .env* and migrations/ flagged as read-only→ runbook .codex/runbooks/RELEASE.md✓ plan written to .codex/memory/HANDOFF.md✓ next session: starts from this plan
Workflow

From blank Codex session to structured delivery.

The same four steps drive every change — onboarding a new repo, shipping a feature, or recovering an incident.

Step 01

Scan the repo

CodexFlow reads project shape, package scripts, framework hints, routes, workflow files, and existing conventions.

Step 02

Install the workflow OS

Generate AGENTS.md, repo memory files, Codex skills, slash commands, runbooks, QA checklists, and handoff templates.

Step 03

Work through structured commands

Plan, implement, debug, review, test, release prep, and handoff — every task starts from known repo context.

Step 04

Keep the repo warm

Update memory, decisions, prompt logs, and handoffs. The next Codex session starts with context instead of confusion.

Scenarios

A day with CodexFlow.

Friday afternoon

Bug fix that ships before the weekend

/plan-feature reads AGENTS.md, scopes the change, hits the QA checklist. /ship leaves a handoff note.

Tuesday

Release prep without panic

Run the release-prep runbook: typecheck, build, security scan, migration check. Codex flags risky files automatically.

New repo onboarding

Onboard a teammate or new project in minutes

CodexFlow scans the codebase, writes PROJECT_CONTEXT.md, and seeds memory so the next session starts with context.

Stop rebuilding context every session. Add the paid CodexFlow workflow to your repo.

Security & delivery

Local-first by design. Signed delivery for paid users.

Trust signals you can verify yourself. Every paid artifact is signed, every install path is auditable, and nothing about your repo leaves your machine.

  • Local-first by designThe CodexFlow CLI operates on local repositories. This release does not transmit usage telemetry.
  • No hidden LLM callsCodexFlow installs workflow files and commands. It never calls model APIs on your behalf in the background.
  • Rights are explicitPersonal and Standard install the same paid payload. The activation file encodes the tier and commercial-use flag.
  • No public paid sourcePaid payload archives are delivered only through secure customer access — not via npm or static assets.
01Signed activation
Ed25519-signed activation file bound to one named individual.
02Signed manifest
Manifest signature verified before the archive is touched.
03Verified archive
SHA-256 + size of the paid payload checked against the manifest.
04Local install
Premium skills and commands emit into your repo — no hidden network calls.

Install once. Improve every Codex session.

Choose a paid license, install the signed payload, and give every Codex session a structured repo workflow.

CodexFlow is independent and is not affiliated with, endorsed by, or sponsored by OpenAI.