PRE-RELEASE — NOT FOR SALE

Status: Foundation in development. This is a decision and coordination framework — files and rules, not an autonomous platform. It does not connect to provider APIs, monitor usage or limits, or run your accounts for you.

What it answers

The questions a multi-tool AI setup keeps asking.

What is AI orchestration?

Treating several AI tools as a team with defined roles, routing, boundaries and handoffs — instead of one chat doing everything.

Why isn’t a single chat enough?

One thread loses context, mixes concerns and has no rules for who does what or when to stop. Coordination needs structure.

How do ChatGPT, Claude and Codex combine?

Each takes the roles it suits — strategy, implementation, or scoped verification — under shared routing and handoff rules.

How is a task routed?

A task is scoped, classified, and sent to the tool, model and reasoning mode that can safely solve it.

How do you choose model and mode?

A decision model maps task type to the smallest reasoning mode that still solves it safely — not the most expensive one by default.

How do you prevent repo conflicts?

Repo gates and worktree rules isolate parallel work so agents don’t overwrite each other.

How do handoffs work?

Verified results are passed forward in structured handoff files so the next agent continues from a known state.

When must an agent stop?

Stop rules define explicit conditions — missing context, scope conflict, risky action — where pausing beats guessing.

What is actually inside · v1.0

The real files you add to your project.

These are the exact filenames that ship in v1.0 — rules, routing and coordination as files, not a hosted app. No file contents are reproduced here, and there is no download yet.

AI Orchestrator Kit Coordinates the AI team.

Entry points

START_HERE_FOR_HUMAN.md Human orientation for the coordination layer.
START_HERE_FOR_AI.md Binding instructions an agent reads before acting.

Roles & routing

AI_ROLE_MATRIX.md Which AI takes which role, with separation of duties.
MODEL_ROUTING.md How to choose a tool and model for a task.
MODE_SELECTION.md The smallest reasoning mode that safely solves it.
TASK_ROUTING.md How a raw task is shaped, scoped and sent.
RESOURCE_POLICY.md Deliberate use of expensive modes.

Repository safety

REPO_GATE.md Pre-change checks that protect the repository.
WORKTREE_POLICY.md Isolating parallel work so agents do not collide.
CONFLICT_PREVENTION.md Keeping parallel agents off the same files.

Coordination & history

HANDOFF_PROTOCOL.md How one agent passes verified work to the next.
REVIEW_AND_VERIFICATION.md How results are reviewed and independently verified.
ESCALATION_RULES.md When to escalate to a stronger model or a human.
STOP_RULES.md When an agent must stop rather than continue.
TASK_HISTORY.md Append-only trail of routed tasks and results.
templates/ Implementation, research, review, QA, handoff and stop-report formats.

Real preview

An example model assignment.

What the routing table can look like once it is filled in for your real tools.

MODEL_ROUTING.md Excerpt

Strategy / sparring ChatGPT Direction, priorities, critical questions
Long-form implementation / complex reasoning Claude Scoped implementation within the project’s rules
Repository execution / tests / verification Codex Gates, checks and independent review

Fictional sample project — no real people or companies.

An example assignment — not a ranking.
Routing is fully adaptable.
No automatic provider control.

What you download

Format & installation.

How the product will arrive — no app, no account, no cloud.

Download format: ZIP archive
Core format: Markdown files and templates
Language of the product files: English — the site is bilingual, the kit files themselves are English
Installation: Copied locally into your project / repository
Account required: No — no cloud, no running online service
Subscription: No — a one-time purchase, no recurring fees or credits
Updates: All v1.x updates included; v2.0 is a separate product
Support: Best-effort email support for installation and reproducible defects

Resource policy

Use the smallest mode that safely solves it.

Reasoning modes are a deliberate choice, not a default. The kit treats model and mode as a configurable decision model — availability and names vary by provider — so expensive modes are reserved for tasks that genuinely need them.

Get early access