Generation System

Dockit's generation system is split into two tracks: the Free template path and the Pro async job path.

Free Pipeline

• Deterministic template generation in the browser
• Fixed delivery profile
• Starter-bundle oriented outputs

Advantages:

• Fast
• Predictable
• No server cost

Pro Pipeline

1. Input Assembly

The web app builds an InfraProjectInput.

2. Policy Checks

The policy engine in packages/shared verifies that the selected combination is valid.

3. Job Creation

The createGenerationJob callable registers the work.

• A cache hit can complete immediately
• A cache miss stores the job in a queued state

4. Job Processing

The server builds the bundle in the following order.

1. Check whether a baseline exists
2. Compute a delta plan
3. Run AI generation or regenerate only changed files
4. Apply fallback artifact repair
5. Run validation
6. Store the result

Required Artifact Repair

The server repairs files that the AI may omit.

• README.md
• .env.example
• docker-compose.yml
• deploy workflow
• flake.nix for specific combinations
• Routing files based on selected options

Cache Strategy

When the input fingerprint matches, Dockit reuses the existing validated bundle.

• Reduces token waste
• Shortens repeated generation time
• Stabilizes quality for identical combinations

Delta Generation

When a baseline input and baseline output exist, Dockit regenerates only the changed files. This is especially useful when:

• Only the database changed
• Only the delivery target changed
• Routing options were toggled on or off

Result Validation

Validation runs at both the bundle and file levels.

• Required file presence
• Artifact combination consistency
• Alignment between requested options and generated files

Why This Matters to Users

The generation system is designed to improve three things that users notice directly.

• Faster repeat generations through cache reuse
• More reliable output through repair and validation
• Smaller update cycles through delta regeneration instead of full regeneration every time
• A clearer handoff from generated output to editable result and export workflow