Workflows

InkHive uses Inngest for durable, observable execution of multi-step AI tasks.

Content generation pipeline

Function: contentPipeline (inngest/functions/content-pipeline.ts) Trigger: content/generate event Config: 3 retries, parallelism enabled

Steps

1. Mark project status as "generating".
2. Generate blog post (must complete first — downstream steps depend on it).
3. In parallel, generate:
   • Social media posts (Twitter, LinkedIn, Facebook, Instagram, Medium)
   • Email newsletter (subject lines, HTML, plain text)
   • SEO metadata (slug, title, description, keywords)
4. On success, set status to "completed".
5. On error, call recordError with the failing step and stack trace.

Inputs

{
  projectId: string;
  inputType: "topic" | "article";
  inputContent: string;
}

Convex mutations called

Failure handling

• Each step is wrapped in step.run with automatic retries.
• If blog generation fails, the pipeline stops — downstream steps cannot run without it.
• If a parallel step (social, email, SEO) fails, other parallel steps continue. The failed step is recorded but does not block the rest.

Publishing pipeline

Function: publishContent (inngest/functions/publish-content.ts) Trigger: content/publish event

Steps

1. Fetch project snapshot from Convex.
2. Run each platform publisher in parallel.
3. Update per-platform publish status via updatePublishStatus.
4. Return consolidated results.

Failure handling

• Platform failures are independent — a Twitter failure does not block LinkedIn publishing.
• Retries are configured at the function level.
• Exhausted retries are logged for observability via onFailure.

Adding a new AI step

1. Create inngest/steps/ai-generation/<your-step>.ts with the prompt and parsing logic.
2. Wire it into contentPipeline or a new Inngest function.
3. Add a Convex mutation to persist the output.
4. Add tests and log token usage.