Product
Everything between the mergeand the post.
Six parts in one workspace. Each works on its own. Together they turn a release into a reviewed, scheduled launch, with a record of every decision along the way.
- Trigger
Release
Recurring invoices · PR #452
- Agent
Launch spec
14 of 14 claims cited
- Agent
Campaign plan
3 angles · 5 channels
- Agent
Visuals
4 on-brand assets · critique passed
- Agent
Copy
X, LinkedIn, Product Hunt, changelog, email
- Human
Review
Approve all 7 posts in one click
- Scheduler
Schedule
Published once, with a receipt
The parts
Six parts.One workspace.
Start with the part you need. Release intake brings the work in, launch intelligence thinks it through, the creative engine makes it visible, approvals keep a person in charge, publishing delivers it, and guardrails hold the whole thing to the truth.
Object model
Structured objects.Not chat output.
Campaigns are database objects, not chat transcripts. Change one hook, regenerate one image, or approve three posts without replaying a conversation. Every generated piece records the versions it came from.
| Object | What it holds | Who changes it |
|---|---|---|
| Release | The brief: title, why it matters, description, audience, and up to twenty source links. | Creators, API keys, webhooks |
| ReleaseSpec | The cited launch spec. Every AI draft, human edit, or applied revision is a new version. | AI drafts, people edit and approve one version |
| CampaignPlan | Channels, angles, assets with capture hints, and the timeline. Versioned like the spec. | AI drafts, people approve |
| CampaignPost | Copy, channel, the claims it relies on, and the spec and plan versions it was written against. | AI drafts, creators edit, approvers decide |
| Design | An editor document with its plan asset, variant label, and critique. | The visuals agent, anyone in the editor |
| ScheduledPost | Destination, time, delivery state, and the provider receipt. | Publishers, then the scheduler |
| AuditLog | Every decision with its actor, request ID, and outcome. | Append-only. Nobody edits it |
Build order
Built inside out.Agents come last.
The platform was built in a deliberate order. Primitives first, so every capability works without a model. Recipes next, so a failure names its step. The agent last, so it chooses among proven capabilities instead of improvising them.
If we reverse this order and start with the agent, we debug hallucinated workflows instead of building a product.
primitivesDeterministic primitives
The editor, brand kits, product surfaces, and campaign objects work on their own, through the UI and the API.recipesNamed, versioned recipes
Control flow is code. Models fill in text at fixed points with schema-checked output, and every step can be retried.agentAn agent on top
The model decides what to do. The application decides how it is done safely, inside one workspace, without approval rights.
Where to next
See it in use.Or see who it is for.
Launch intelligence is in early access for design partners.