Code doesn't just write itself.It orchestrates.
The AI code companion engineered for compounding teams. Codex plans, writes, reviews and ships — across every repository, runtime, and agent in your stack.
One canvas. Every agent in your stack.
Switch between plans, edits, reviews, and ships without losing context. Codex preserves intent across every surface.
1import { codex } from '@codex/core';2// Codex composes the agent graph for you4export const pipeline = codex.graph({5plan: agents.planner({ depth: 'deep' }),6write: agents.coder({ model: 'omni-x' }),7review: agents.reviewer({ strict: true }),8ship: agents.deployer({ canary: 0.1 }),9});
C·O·D·E·X
A single coherent loop — plan, write, review, ship — across every repository, runtime and agent in your stack.
NODE / 00.41.07Numbers that compound, silently.
Measured across 4,200 engineering teams shipping with Codex.
Between
commits.
Codex doesn't just autocomplete. It maintains a living memory of your intent — every PR, every diff, every architectural decision — and keeps the entire team aligned on the next move.
Six surfaces. One coherent loop.
Codex traces every dependency before it writes a line.
A live structural graph keeps refactors safe across hundreds of files. Hover to watch the path.
Speaks every framework in your monorepo.
Codex ships with native adapters for the languages and runtimes your team already trusts.
Shared cursors. Shared intent.
Live presence, conflict-free edits, and audited agent runs — across every contributor.
Edge-routed agents. 240ms p95.
Workloads route to the nearest GPU pool — automatically.
Right model for the right task.
- planomni-x · large
- editcodex-fast
- reviewsentinel-pro
- embedatlas-mini
A tiny DSL with a massive surface.
Describe outcomes, not steps. Codex compiles your intent into a deterministic agent graph you can version and review.
// orchestrator.codex
task "refactor billing engine" {
plan :: deep, ttl=10m
budget :: 80k tokens
guard :: tests.green && lint.clean
ship :: pr → review → canary(10%) → 100%
}Built for one dev.
Priced for every team.
Flat per-seat pricing, no token surprises. Scale orchestration without scaling spend.
Hobby
For solo devs exploring agent-driven workflows.
- 1 workspace
- 200 agent runs / mo
- Community support
- Local-only repos
Team
PopularFor compounding engineering teams shipping daily.
- Unlimited workspaces
- 10k agent runs / mo
- Parallel orchestration
- PR reviews & policies
- SSO + audit log
Enterprise
For platforms with strict security & scale needs.
- Dedicated tenant
- Bring-your-own model
- SOC2 + HIPAA
- Private VPC routing
- 24/7 white-glove
Documentation that moves at your speed.
Getting started
01Core concepts
02Shipped weekly. Compounded daily.
- v3.4.0May 28, 2026Major
Parallel agent orchestration
Spawn up to 12 agents in a single plan with deterministic merge. New scheduler routes tasks by repo, runtime, and historical token cost.
- v3.3.1May 14, 2026Improvement
Diff-aware code review
Reviews now respect existing PR conventions and learn from squash-merge history. Up to 3× fewer false positives in monorepos.
- v3.3.0Apr 30, 2026Feature
Bring your own model
Route any model behind a single Codex endpoint — OpenAI, Anthropic, Mistral, self-hosted. Policies enforced per workspace.
- v3.2.4Apr 12, 2026Fix
Editor sync, faster
VS Code sidebar now streams agent thoughts under 60ms p95. JetBrains parity coming next month.
Four commands.
Infinite leverage.
Drop Codex into any repo in under a minute. No infra, no rewiring — just orchestration.
Join the orchestration era.
Free to start. Connect a repo and ship your first agent-run pull request in under five minutes.