GridWork HQ
AI Pipelines

AI Pipelines

17 built-in AI pipelines for lead generation, client delivery, and agency operations.

AI Pipelines

GridWork HQ includes 17 AI pipelines that automate agency workflows end-to-end. Each pipeline is a Markdown definition file that the pipeline server reads at runtime — no code changes are needed to add or modify pipelines.

How Pipelines Work

  1. You trigger a pipeline from Mission Control, the dashboard API, or a cron schedule
  2. The pipeline server reads the pipeline definition from the knowledge vault (system/rules/pipelines/)
  3. It spawns a Claude Code CLI process with the pipeline instructions as the system prompt
  4. Claude executes the steps: reading files, calling tools, generating output
  5. Results stream back to the dashboard via SSE and are saved to the .output/ directory

Pipeline Categories

Lead Generation

PipelineTriggerDescription
Prospectprospect [type] in [area]Find and score local business leads
Outreachoutreach [business]Draft personalized outreach emails
Auditaudit [business]Free external site audit for prospects

Client Delivery

PipelineTriggerDescription
Proposepropose [business]Generate proposals from completed briefs
Specspec [client]Requirements, design, and task breakdown
Planplan [business]Create detailed site build plans
Buildbuild [business]Execute build plans across 5 phases
Brandbrand [business]Generate full brand identity systems
SEOseo [business]Post-launch SEO analysis and optimization
Contentcontent [business]Generate a month of blog posts, social, and email content
Reportreport [business]Monthly maintenance reports for retainer clients
Follow-Upfollow-up [client]30-day post-launch check-in with retainer pitch

Operations

PipelineTriggerDescription
Internal Auditinternal-audit [site]Deep 10-category technical QA
KB Librariankb-librarianKnowledge vault audit and maintenance
Archive Outputsarchive-outputsArchive old output files
Friday UpdateCron onlyWeekly client update emails
Scope AuditCron onlyDetect scope creep against build plans

Model Routing

Each pipeline specifies its model tier in YAML frontmatter. Multi-pass pipelines use different models per pass:

ModelTypical Use
HaikuLightweight maintenance (kb-librarian, archive-outputs)
SonnetImplementation, drafting, technical analysis
OpusPlanning, review gates, high-quality final output

Multi-Pass Pipelines

Several pipelines use multiple passes with different models. For example, the audit pipeline uses Sonnet for technical analysis (Pass 1) and Opus for sales framing (Pass 2). The spawner runs each pass sequentially, with intermediate results saved to .output/.

Adding Custom Pipelines

See the Advanced Extensions guide for instructions on creating new pipeline definitions.

On this page