Beta
GTM Harness
Your go-to-market, run by agents you can check, with any model: a marketing brain, a skill per role, reviewed drafts, approvals from your phone and one-tap sends.
The split
- LLM writes
- The plan, the research, the campaign directions and the first messages, with any model
- System One decides
- Whether each draft is ready, needs a revision, or is blocked
- Code executes
- Validation, the scorecard math, the workspace, approvals bound to the text, opt-outs, the send links and the evals
Engine
Any model writes the plan and the drafts (Claude, or any provider through Vercel AI Gateway, OpenRouter or a /chat/completions endpoint); the brain reviews every draft (ready, revise or blocked); code packs the workspace, binds each approval to the exact text, enforces opt-outs and builds the one-tap send links. The founder sends everything.
Rulebook
What correct means here, and the mistakes we've already caught. It grows one job at a time.
- R01The harness never sends anything; the founder does.
- R02No invented traction, numbers, customers or quotes, in the plan or in the drafts.
- R03The reviewer can block a draft but never ship one.
- R04Every edit the founder makes goes in the corrections log; repeated ones become rules.
- R05An approval covers the exact text it was given for; an edited draft needs approving again.
- R06Anyone marked do_not_contact gets no approval card and no send link, whatever was approved before.
- R07The first WhatsApp message to anyone is the founder's own tap; the WhatsApp Business API is only for people who opted in, and outside the 24-hour reply window it sends only approved templates (research/gtm-harnesses.md).
- R08No bot logs into anyone's WhatsApp, Telegram or LinkedIn, and research reads logged-out public pages only, with no stealth tools or CAPTCHA solving (research/gtm-harnesses.md).
- R09A model is called supported only after it passes the harness evals on every run.
Intake
- Your product in one or two sentences
- Who it's for, and your goal for the month
- The channels you can use
- UNIT
- One harness for one product
- REVIEW LAYER
- Every draft carries the reviewer's verdict and fixes; the founder approves each message, and accepts new rules each Monday.
- DELIVERY
- A web run in about a minute, and a workspace any agent can run (Claude Code, Codex, Cursor, Copilot, Gemini CLI): AGENTS.md, the brain, workflows, 11 skills, a first campaign, reviewed drafts, the pipeline, the corrections log, the sprint and the dashboard. The `pnpm gtm` CLI adds Telegram approvals and send links.
- TURNAROUND
- About a minute
How we find buyers
- Founder communities and office hours
- Startup programs and accelerators
- X threads with a live demo