← All products
🖥️

AI Agent Interface Kit

Your AI agent works in text. You think in visuals. This kit bridges the gap with Notion, Linear, Fizzy, or any REST API, a dispatcher shim, a two-board architecture, and now a notes-inbox capture layer so raw ideas do not pollute the real task board.

✓Architecture patterns from 54 commits, 21 bug fixes, and a one-day migration across 50+ automation scripts.
Updated May 11, 2026 · Added notes-app inbox capture pattern and corrected dispatcher/hook paths in the setup docs.

Paid Digital Thoughts subscriber?

Yearly subscribers: all products free ($1150.94+ value)

Monthly subscribers: 1 free product per month — subscribe from $29.99/mo

Claim your free copy →
$49.99One-time purchase. Lifetime updates.

What changed in this update

  • ✓Added notes-inbox-capture.md
  • ✓Added Fizzy integration notes
  • ✓Corrected dispatcher and hook paths in setup docs

Best for

  • +People running agents across CLI, chat, cron jobs, and webhooks
  • +Builders who need visibility without building a whole task product

Not for

  • -A polished SaaS dashboard you can use without any setup

What you get

+Dual-mode: knowledge for you, executable scripts for your agent. Hand the ZIP to any AI agent and it sets up the whole system
+Three integration paths: plug into Notion/Linear via API, fork an open-source kanban, or build custom
+Dispatcher shim pattern: swap backends with one env var, zero consumer script rewrites
+Two-board architecture: Ops board (your tasks) + Automations board (agent definitions that never close)
+Notes inbox capture: quick notes get triaged before they become cards
+Tag namespacing: replace 5 database tables with one concept (area/X, p/High, src/cli)
+Session lifecycle: one card per CLI session, subtasks as checklist steps, PID-based ownership
+Automation lifecycle: Intake to Idle to Running to Needs Attention, with comment gating for high-frequency runs
+Natural-language intake: drop a card like "Send me weather at 7am", agent converts to working automation
+Works with ANY AI agent: Claude Code, GPT, Gemini, Cursor, open-source, custom

Package includes

  • •Comprehensive guide (10 chapters, 10,000+ words): architecture, patterns, migration, rollback
  • •AGENT-INSTRUCTIONS.md: self-contained prompt any AI agent can read and execute to build the system
  • •setup.sh: interactive bootstrap that configures everything for your chosen backend
  • •DECISION-GUIDE.md: 5-question framework to pick the right path for your situation
  • •templates/dispatcher-shim/dispatcher.py: reference implementation with abstract backend interface
  • •4 backend adapters: Notion API, Linear GraphQL, Fizzy/Rails, Generic REST (with Trello example)
  • •templates/hooks/: session-start.sh and session-end.sh for CLI integration
  • •templates/automation-runner/: drift-tolerant automation scheduler with lifecycle management
  • •notes-inbox-capture.md: fast capture layer before board promotion
  • •Board setup configs: column definitions and tag scheme for both Ops and Automations boards
  • •Migration checklist: pre/during/post migration steps from real production experience
  • •Cheatsheet: single-page quick reference for commands, lifecycles, and common fixes
  • •Platform-specific setup guides for Notion and Linear
  • •CHANGELOG.md

FAQ

Does this only work with Claude Code?

No. Every template uses generic environment variables (AGENT_BOARD_BACKEND, NOTION_API_KEY, etc.). The AGENT-INSTRUCTIONS.md works with any AI agent that can read markdown and run shell scripts.

Do I need to deploy Fizzy?

No. The kit supports three paths. The easiest is plugging into Notion or Linear via their APIs. You only deploy Fizzy (or another open-source kanban) if you want full control. The DECISION-GUIDE helps you choose.

What if I already use Notion/Linear for tasks?

Perfect. The kit includes backend adapters for both. Run setup.sh, pick your backend, enter your API key, and your agent starts creating cards on your existing board within minutes.

How is this different from the AI Agent Blueprint?

The Blueprint builds the agent itself (memory, skills, identity). This kit builds the visual interface layer on top. They complement each other. The Blueprint is about what the agent knows and does. This is about what you see.

Can I feed this to my AI agent?

Yes. That is the primary use case. AGENT-INSTRUCTIONS.md is a self-contained prompt. Your agent reads it, asks which backend you want, runs setup.sh, and builds the whole interface layer. You watch cards appear on your board.

Secure checkout by Stripe. Instant download + guided Claude Code setup.