Playbooks

Playbooks

Define multi-step AI workflows in markdown and run them with a single click. Playbooks chain AI steps together with branching logic, variable passing, and human-in-the-loop controls — all without writing code.

What’s new

• Markdown-defined workflows — Write playbooks in a structured markdown format with ## STEP sections, ## INPUTS blocks for variables, ## BRANCH blocks for IF/THEN/ELSE logic, and ## ARTIFACTS for structured output.
• Step chaining — Pass output from one step to the next with @output(varname) directives. Extract specific fields from AI responses with @output(varname, extract:"field").
• Branching logic — Route execution based on step outputs using ## BRANCH blocks. Supports conditional paths with IF/THEN/ELSE evaluation.
• Human-in-the-loop — The @elicit(type, "prompt") directive pauses execution to collect user input mid-workflow. Respond inline and resume.
• Breakpoints — Pause execution at specific steps, inspect and override variables, then resume. Useful for debugging and oversight of complex workflows.
• Streaming execution — Watch AI responses arrive token by token for each step. The execution UI shows a 3-column layout: step list on the left, streaming output in the center, and a variable tracker on the right.
• CodeMirror 6 editor — Custom playbook grammar with syntax highlighting (color-coded steps, sections, branches, variables, prompts, outputs, and elicit directives). Includes a snippet toolbar for inserting common blocks and a @ popunder for searching and referencing prompts from your library.
• Flow graph visualization — The playbook detail page renders a visual flow graph showing system prompts, inputs, steps, branch points (with diverge/converge paths), and artifact nodes.
• Prompt references — Reference prompts from your library within playbook steps. The editor provides @ search for quick lookup.
• Artifact system — Define an ## ARTIFACTS section to produce structured output in markdown, JSON, Mermaid diagrams, Chart.js, HTML/CSS, JavaScript, or TypeScript. Artifacts are downloadable from the results page.
• Restart from step — Re-run a completed execution starting from any step. The new execution tracks its parent for audit history.
• Per-step cost tracking — Token counts, latency, model, and cost estimates stored for each step and displayed in results.
• 8 starter playbooks — Pre-built templates seeded on first use: Code Review Pipeline, Content Brief Generator, Research Synthesis, Technical Decision Matrix, Multi-Audience Content Adapter, Full Stack Feature Spec, Interview Prep Coach, and Incident Postmortem Generator.

Output destinations

Route completed playbook results to external systems automatically:

• Webhook — POST results to any HTTPS endpoint.
• Email — Send results to your verified email address.
• GitHub — Push results to a GitHub repository.

Deliveries retry up to 3 times with exponential backoff if the destination is temporarily unavailable.

Trigger URLs

Start playbook executions from external systems via unique trigger URLs. Each trigger gets a cryptographic token and configurable rate limiting (requests per hour). When a triggered execution pauses for human input, you receive an email notification with a direct link to respond.

Persistent hibernation

Paused playbook runs survive server restarts. Hibernated state is stored in your database and restored when you resume or respond to an elicitation — no time limit on how long a run can stay paused.

For API integrators

Thirteen MCP tools for playbook management and execution: