Auteur Intelligence™
Lineage artwork for Auteur works with Codex
Lineage visual · integration schematic codex

Harness integration

Codex

Auteur works with Codex

Connect Codex to Auteur and keep working the way you do now, with persistent projects, governed campaigns, verification, and receipts underneath. The agent can change; the project, process, authority, and evidence remain.

What Codex does well

Codex is a capable coding agent with MCP support and a headless mode that emits a machine-readable stream. It handles long, scoped engineering work well, which is why Auteur both connects to it as a client and drives it as a campaign seat.

What it brings to the seat

Codex is the workhorse: the single most-used executor in our own build ledger, at home as a builder inside campaigns or as the orchestrator running them. House it on the subscription you already pay for through its own sign-in, or connect an API key. It pairs naturally with a different model family reviewing its diffs, which is exactly how the campaign system seats it.

“Codex, compose the narration for chapter two in Henry’s voice and lay it against the picture.”

A thing no coding agent does alone

“Drafted the read from canon, generated it through your ElevenLabs account, placed it on narration lane A2, and staged the timeline to the Resolve inbox. The receipt lists 14 clips and $0.86 of voice spend on your plan.”

Illustrative exchange. Canon-grounded narration, lane placement, and the Resolve inbox are shipped surfaces.

What Auteur adds

For someone already using Codex, Auteur adds a persistent project that keeps its own canon and evidence instead of losing them to the session; governed multi-round campaigns with acceptance gates and recovery; production tools beyond code; and canonical receipts that verify what actually changed independently of the agent's own claim. Codex stays your interface; Auteur is the amplification and accountability layer behind it.

How it connects

Which Auteur products it reaches

As a client, Codex can address the Auteur domains you are entitled to (writing and story, prep, film, game, learning, and the Build campaign system) in natural language. As an executor it runs Build campaign seats. Reachability is not entitlement: the connection can address these domains; which you operate depends on what you hold.

What is measurable

When Codex runs as an Auteur Build executor or through the direct-model lane, Auteur derives token and cost accounting from the record of the work it ran. What you do in Codex outside an Auteur campaign is not visible to Auteur.

Auteur never invents missing usage data; missing = unknown, never zero. No token or cost figure is ever shown for work performed outside Auteur's visibility.

What this integration does not do

Security and authority boundary

Connecting Codex does not grant it wildcard authority over your files, spending, publication, or application changes. The connection is an interface, not a grant of authority. What a connected agent may do stays governed by Auteur, not widened by connecting it.

Current status

Current per-mode status is shown above.

Proof

Codex is the most-used builder in our own ledger: 1,884 receipted tries ran through Codex seats (measured from Algernon's work-order receipts, 2026-11-05). Every worker's record, Codex included, with what it cost and how often it finished, is on the build campaign proof page.

Set it up

To operate Auteur from Codex as a client, the invited local Conductor connection is the supported route:

Request Invitation. A small, deliberately limited cohort. Asking starts a conversation with us. Founding seats are limited and offered first come, first served.

To run Codex as a Build executor inside a campaign, start with Get founding access.

Affiliation

Auteur works with, and is designed to connect with, Codex. No partnership, endorsement, or certification is claimed, and none exists.

Powered by the Algernon Engine, in every product

◈

Memory

The project remembers what the chat window forgets, approved canon kept separate from suggestions, on every surface, under any model. It travels with the project, not the app.

Go deeper → ⊚

Eval

Work is judged, scored, and receipted instead of accepted on vibes, evidence before acceptance.

Go deeper → ⌘

Workshop

Hard problems, on demand: two or three of your own models attempt it within a cap you set, a judge explains its pick, and you decide what enters the project.

Go deeper →
Per-mode registry status
client in use internally certification pending executor in use internally certification pending direct model available for measured accounting

Auteur provides the harness, the scaffolding, and the software engineering · not the models' meters. We can evaluate, and help you set up, local open-source models; your providers connect as seats. Model access is yours: pay-as-you-go accounts such as OpenRouter or Wavespeed, the subscriptions you already hold, or credits at the provider's list price. We don't mark up AI usage.