Welcome — I run the team here. Every conversation in the sidebar is a problem we solve, top to bottom by how much it matters. Start with yours, or just tell me what brought you in.
Tell your AI team the goal.
Approve the plan. Watch it ship.
Action Plan gives you a full team of specialist AI agents — strategist, developer, researcher, assistant, analyst — that plan, execute, and review real work in the background, and bring you only the decisions that need a human. Local-first. Budgeted. Fully audited.
Action Plan is in private preview. This message joins you to the waitlist — when onboarding opens, your first conversation starts right here.
Got it — you're on the early-access list, and I've saved your goal: . When your invite opens, this is the first thing we'll work on together. Watch your inbox.
How it works
The whole loop: goal → plan → approval → governed execution → evidence → done.
Say the goal
"Build the booking app my shop needs." "Automate my email." "Ship this feature." Your Orchestrator turns it into a concrete plan with scope, steps, and cost.
Approve the plan
You see exactly what will run, what it may cost, and which agent does what. Adjust anything — then approve once.
Review the work
Agents execute in the background, behind quality gates. You get evidence-backed results and only the decisions that truly need you.
A team, not a chatbot
Every agent has a stable role, a personality, real memory, and a lane. They work together — propose, challenge, review — and escalate to you only when judgment is needed.
…and 17 more specialists: Rapid Builder, UX Director, Sales Expert, Marketing Lead, Startup Counsel, Health Coach, Habit Partner, Meeting Leader, Customer Experience Lead, Truth & Bias Analyst, and more.
What your team can do
Outcomes, not features — each one is its own conversation in the sidebar if you want the details.
Work that advances itself
Background agents move your projects forward — planning, building, reviewing, documenting — and stop at exactly the moments that need a human decision.
- Multi-step workflows run end-to-end in the background
- Parallel agents fan out across independent work
- Everything that needs you lands in one review queue
Approve with evidence, not vibes
Every proposal arrives with the diff, test results, reasoning, and history attached. Approve, edit, or reject in one click — every decision is recorded.
- Evidence-first review cards: what changed, why, what it cost
- Quality gates run before work ever reaches you
- A durable audit trail of who approved what, and when
Know the cost before it runs
Action Plan forecasts what a job will cost — in tokens and dollars — before spending anything, enforces budgets, and tracks every model call in one ledger.
- Pre-run risk-and-cost forecasts gate big jobs
- Budgets per project, per workflow, per agent
- One ledger across local models and subscriptions
A memory you can trust
Agents remember your preferences, projects, and decisions — in a versioned memory ledger where every change is proposed, reviewed, and reversible.
- Agent-proposed memory edits require approval
- Contradictions escalate to you, never silently overwritten
- Memory effectiveness is benchmarked, not assumed
Your knowledge, on tap
Ingest your codebase, documents, books, notes, and web sources into governed knowledgebases. Agents cite what they retrieve — answers trace back to your sources.
- Code, books, notes, and architecture knowledgebases
- Semantic + lexical + graph retrieval with citations
- Traceability from idea to shipped change
Workflows you own
Workflows, prompts, rules, and checklists are plain files you can read, edit, and version. Start from a deep catalog; the agents follow your process.
- A growing catalog of business and delivery workflows
- Every workflow editable — no black boxes
- Scheduled and recurring runs: standups, reviews, reports
A real workspace, not just a chat window
Threaded multi-agent chat, projects and tasks, a block-based notes editor with backlinks, an AI-scheduled calendar, and live operations dashboards.
- Threaded chat with mentions, reactions, and proposals
- Projects, tasks, and review queues agents keep current
- Notes, calendar, and knowledge in the same frame
Local-first, secure by construction
Runs on your computer. No required cloud. A built-in security program — scanning, least-privilege tool access, prompt-injection defenses — from day one.
- Your data stays on your machine by default
- Per-workflow least-privilege permissions
- SOC 2-aligned controls built in, not bolted on
Run it your way
Mix all three. The router picks per task; the ledger keeps score.
On your computer
Download and run locally. Your files, your repos, your data — nothing leaves your machine unless you connect it.
With local models
Run open models via Ollama for private, zero-marginal-cost work. Action Plan routes each task to the cheapest model that can do it well.
With your subscriptions
Bring your Claude or OpenAI subscription or API keys. The cost ledger tracks everything — and shows what local routing saved you.
Honest comparisons — because you'll make them anyway
Most AI tools are excellent at one slice of this. You've probably felt where each one stops. Action Plan's bet is the governed whole: a full team, your whole workload, with budgets, audit, and your approval in the loop.
If you use an AI chat assistant
ChatGPT · Claude · Gemini
Answering, drafting, brainstorming — on demand, when you ask.
- It waits for you — nothing happens unless you're typing
- It forgets between sessions, or remembers the wrong things
- You get answers, then YOU still do the work
- No budgets, no approvals, no record of what it did
- Agents hold your goals and work them in the background, on schedule or on demand
- An approved, versioned memory that survives every session
- You receive finished, evidence-backed work — not homework
If you use a coding agent
Cursor · Copilot · Devin · Goose
Autonomous code generation with very little friction.
- Engineering-scoped: your email, projects, finances, and research are someone else's problem
- Ungoverned: no approval lineage, no spend forecast, no audit trail
- One agent or an anonymous swarm — no roles, no team, no accountability
- The same vibe-coding loop — describe the app, review the plan, ship — but governed: cost forecast first, quality gates, evidence at review
- And the rest of your work runs through the same team: product, research, ops, finance, life
- A planning layer (goals, timelines, risks, reviews) no coding agent has
If you use an automation platform
Zapier · n8n · Make
Deterministic trigger-action plumbing between SaaS apps.
- Recipes execute; they don't plan, reason, review, or improve
- Every edge case is a new zap; brittle chains nobody remembers building
- Your workflow logic lives in someone else's cloud, in a diagram you can't diff
- Agents design the automation with you, run it, verify the result, and propose improvements
- Workflows are readable files you own and version — not locked diagrams
- When something breaks, an agent diagnoses it and brings you the fix to approve
If you build on an agent framework
CrewAI · AutoGen · LangChain · Letta
Letting developers assemble custom agent systems from parts.
- You're building the product yourself: memory, budgets, review UX, audit — all DIY
- Agent memory is typically self-edited and unaudited; hallucinations go unmeasured
- Months of glue code before the first useful outcome
- The team, memory governance, budgets, security program, knowledgebases, and review UX are already built
- Memory writes require approval, reads are logged, fabrication is a measured metric — no surveyed framework does all three
- Useful on day one; extensible when you want it to be
| Capability | Chat assistants | Coding agents | Automation platforms | Agent frameworks | Action Plan |
|---|---|---|---|---|---|
| Named multi-agent team with roles & memory | — | — | — | DIY | ✓ |
| Background execution toward your goals | — | code only | recipes only | DIY | ✓ |
| Approval lineage + full audit trail | — | — | partial | — | ✓ |
| Cost forecast before running + budgets | — | — | — | — | ✓ |
| Governed, versioned, measured memory | partial | — | — | partial | ✓ |
| Your documents/books/code as cited knowledge | partial | code only | — | DIY | ✓ |
| Planning layer: goals, timelines, risks, reviews | — | — | — | — | ✓ |
| Local-first: your machine, your data | — | partial | — | partial | ✓ |
| Security program built in (SOC 2-aligned) | vendor-side | — | vendor-side | — | ✓ |
| Editable, file-based workflows you own | — | partial | proprietary | code | ✓ |
Keep the tools you already use
Action Plan isn't asking you to rip anything out. It's designed to sit alongside your existing stack as the task and project management layer that coordinates it — the one place where work is planned, budgeted, approved, and tracked, whatever tool executes it. Your chat assistant, your coding agent, your zaps: Action Plan fills the gaps between them and owns the follow-through.
And when a connection you need doesn't exist yet? You'll be able to ask your team to build it — describe the integration, and your agents develop, test, and propose the connector for your approval. Easy custom company integrations, generated by Action Plan itself, are on our near-term roadmap.
What happens after you join
1. You're on the list
We record your email and your goal. No spam — only launch news and your invite.
2. We read every message
Your "what do you want to automate" note directly shapes onboarding paths and which integrations we build first.
3. Early access rolls out in waves
Invites go out as capacity opens. A concrete use case moves you up.
4. Your first conversation is ready
When you get access, onboarding starts from exactly the prompt you typed here.
Questions, answered
What is Action Plan, in one sentence?
Will it work with the tools I already use?
When can I use it?
Does my data leave my computer?
What does it cost to run?
Do I need to be technical?
How is this different from ChatGPT / Copilot / Zapier?
What happens to what I type in the box up there?
Your team is almost ready.
Tell us what you want to automate, and be first in when the doors open.