Auteur Conductor

Bring the AI harness you already use; Auteur adds persistent projects, governed campaigns, production tools, verification, and receipts. Conductor is a flagship feature of the platform, not a separate product: it comes with every Auteur product, and it is the harness-agnostic control surface that lets a compatible external AI client discover and operate the Auteur domains you have authorized, in natural language.

The patchbay case, with cables running into every other case: your harness is the console you drive from, and Auteur is what the patches actually reach.

Bring your own harness

The Auteurs: Powers #2, The Conductor

Your coding agent forgot what your writer decided. Auteur Conductor plugs the AI agent you already use into one shared memory and one set of tools, and anything that spends or publishes still waits for you.

Transcript

Your AI team is full of geniuses… who don't talk to each other. Your coding agent forgot what your writer decided. The research went nowhere. Welcome to the Tower of Babble. Enter… the Conductor. Auteur Conductor plugs the AI agent you already use into one shared memory… and one set of tools. Claude Code. Codex. Cursor. Antigravity. Grok. And for builders, Copilot's coding tools, through MCP. One wave of the baton… and the right model takes the job. One shortcut drafts your posts everywhere. Another finds the writers for your manuscript. Another starts the research. Every move lands on the record… and anything that spends or publishes still waits for you. Different geniuses. One orchestra. Auteur Intelligence. Conductor included. Be your own auteur.

Captions (.vtt)

Change AI interfaces without losing the project, process, approvals, or evidence. Conductor supports configuration for the harnesses professionals already use (Claude Code, Codex, Cursor, generic stdio MCP, and generic JSON clients) while preserving one server authority. Per-harness compatibility is published for each named client rather than as a blanket claim.

Within that authority boundary, work stays inside approved authority, review gates remain visible, and unwired verbs are removed rather than left to fail quietly.

“Coordinate several agents across two computers, pause at every approval gate, and resume from the last verified state.”

One arsenal: models, providers, harnesses, and local compute

Conductor lets a supported harness take advantage of supported models from the providers you already hold, by subscription, OAuth, CLI, or API, and of local compute you own. They work together in one project, with real benchmark receipts and evaluation, and long-term memory of what worked and what did not, regardless of platform. Because the mixture spans providers, not just models, the mistakes of one lab get caught by the instincts of another.

The conductor's judgment

Bringing many models is only useful if something intelligent decides between them, and that is the layer chat harnesses do not have.

Capability-aware routing: the studio knows what each seat can and cannot do (which handle video, which handle images, which carry the long context a big build needs) and routes the work accordingly instead of letting a text-only seat fail at a video job.
Build-time forecasts: campaigns estimate remaining time from prior history, and the forecast updates as rounds land, so you know what an overnight run really means before you start it.
Stop, switch, resume: when a seat is failing, the build stops before it burns more spend; the work reseats to a different model and resumes from the receipts, not from zero. Fully automatic mid-flight reseating is where this is designed to go.
Failures remembered: every failure is classified and recorded to the model ledger, so the same mistake is not bought twice: the next build routes around what already went wrong.

“Seat Codex as the builder and Claude as the reviewer, give them separate worktrees, and wake me only if a gate fails twice.”

How it works

Your harness is the console; the authority stays on the Auteur server. Conductor keeps the tool list scoped to what a job needs instead of flooding the session, and ordinary-language requests reach bounded domain actions across writing, prep, visual work, Resolve, and campaigns.

Dynamic domain packs: a small discovery core that activates task-relevant capabilities instead of exposing the whole platform every turn.
Call-time authorization: server-side permission checks applied to each request.
Natural-language production control: ordinary-language requests mapped to bounded, reviewable domain actions.

Visibility is not authority

Connecting an AI interface does not grant wildcard authority over files, spending, publication, or application changes. A tool can stay visible in a stale client list, but authorization is decided at call time, not at connection time, so a connected tool is limited to the authority it currently holds.

Where it runs

Conductor runs beside the harness you already use and connects it to your Auteur workspace as a local connection. The hosted remote endpoint is not deployed yet; with the hosted workspace, the connection is read-only monitoring at launch, and Conductor actions follow.

By 2026-08-28, the local implementation had been tested, and natural-language production control had run across several domains, not yet all of them.

Every harness at parity

No harness is featured over its peers. Each is built to connect through the same server authority and carries its own per-mode compatibility state. The integrations hub carries the current state for every named harness, each on its own page. As an editorial rule, this site does not present a usage figure a provider did not report.

Questions

Do I buy Conductor separately? No. Conductor is a platform feature that comes with every Auteur product; see all products and pricing.

Do I have to move off the tools I already use? No, that is the point. Auteur is built to let you keep working in your existing harness, with your project, approvals, and evidence coming with you.

Take the next step

Get founding access. Conductor comes with every product.