oneshot-gtm

oneshot-gtm

Open-source GTM agent for technical founders. Pay-per-result, signed receipts, founder-led discipline encoded. Terminal CLI + local web dashboard over one SQLite ledger.

oneshot-gtm.com · what a signed receipt is · docs

bunx oneshot-gtm-server     # dashboard only — published, no clone

https://github.com/user-attachments/assets/bba2fb2d-35c3-4171-a358-fd3a987c24bc

---

What this is

OneShot is a pay-per-use API toolbox — email, SMS, voice, deep research, person enrichment, browser automation, website build — settled per call in USDC on Base, with a cryptographically signed receipt for every action.

oneshot-gtm is the strategy wrapper. It encodes the founder-led-sales playbook — Mom Test, Sean Ellis 40%, Predictable Revenue, do-things-that-don't-scale, multichannel cadence, signed-receipt CAC — as named plays you run from the terminal or the dashboard. Finders watch public signals, an ICP gate filters what they find, you approve rows in a queue, plays draft and send, cadences follow up, and every dollar lands in a receipt you can attribute to an outcome.

MIT, so you can read every prompt, fork every play, and trust what's running.

Why not Apollo / Clay / Outreach / Smartlead

	Them	oneshot-gtm
Pricing	Seat-based SaaS, $$/seat/mo	Pay-per-result, no subscription
Source visibility	Closed; you trust the dashboard	MIT; read the prompts, fork the plays
CAC story	Blended, estimated, dashboard-shaped	Signed per-call receipts, exportable
PMF posture	Assumes PMF, optimizes sends	Pre-PMF aware, soft-gates on scale moves
LLM	Built-in, opaque	BYO key (OpenRouter / OpenAI / Anthropic)
State	Vendor cloud	Local SQLite + chmod-600 dotfile

Most GTM tools assume you have product-market fit and optimize sends. Most pre-PMF founders don't, and end up scaling a broken motion because the tool said "send more" — the Startup Genome Report's top documented cause of startup death. So the discipline is built in:

• Plays default to founder-to-founder voice, low volume (≤50/day), one touch unless you invoke the cadence engine.
• Every first touch is Hook → Identity → Offer → CTA. The Offer says the useful thing in the email, for free; the CTA asks for one line the reader can answer from their own experience, or asks for nothing. It never asks a stranger for a meeting. One true concession you write in config (founderAdmission) is worked into roughly a third of first touches; leave it blank and the beat is skipped, never invented.
• Every draft passes a lint built on the Wikipedia "Signs of AI writing" canon — banned phrases, em dashes, AI vocabulary, three-item lists, sycophantic openers.
• Scale-move commands (handoff templatize, first-ae, readiness) print soft-gate checklists and default to "not yet, fix this first" until the signals earn the move. --force overrides.
• Every paid action emits a signed receipt carrying a memo (why the call happened), structured decisionContext, and a goalId grouping a cadence's spend. When a reply or deal outcome lands, that value is tagged back, so CAC and RoCS on the Measure page are attestable and outcome-attributed, not estimated.

---

Setup

curl -fsSL https://bun.sh/install | bash        # Bun is the required runtime

git clone https://github.com/oneshot-agent/oneshot-gtm
cd oneshot-gtm && bun install

bun run cli -- init                             # config + keys wizard
bun run cli -- doctor                           # sanity check
bun run --cwd apps/web build                    # one-time: build the SPA
bun run cli -- ui                               # http://127.0.0.1:3030

init asks for the founder profile the prompts draw on: background that builds trust, products you've shipped, notable partners or customers, and one true concession. All optional — a blank field skips the beat that uses it rather than improvising. Edit any of them later from /setup or config founder. What reaches the model, and when maps each field to the surfaces it shapes. A voice card adds how you write, in your words, to every email draft.

Some finders need keys init never asks about — most people want GITHUB_TOKEN (a classic token with no scopes) before enabling the GitHub finders. /setup and config keys store them; finders lists them.

To call it from anywhere: cd apps/cli && bun link && bun link oneshot-gtm && cd -. Prefer the dashboard without cloning? bunx oneshot-gtm-server downloads and boots it. Bun is still required — the bundle uses bun:sqlite and Bun.serve, and fails loudly under plain node. The CLI itself is not published to npm.

---

The two surfaces

Both read and write the same ~/.oneshot-gtm/ledger.sqlite.

Terminal

bun run cli -- intel advise                        # interactive coach
bun run cli -- find watch --once                   # poll due triggers, enqueue candidates
bun run cli -- find drain podcast-guest --dry-run  # preview approved /queue rows
bun run cli -- cadence advance                     # daily tick: poll inbox, fire follow-ups

73 commands — bun run cli -- --help (or oneshot-gtm --help once linked) is the reference:

Group	Commands
init · doctor · ui	setup wizard · health check · open the dashboard
config	llm · founder · keys · voice [--from ...] · linkedin-session · telemetry on|off · slack-webhook [url] · x-engine [engine] · spend-ceiling [amount|off]
identities	list · add · remove — the sender pool
gmail	auth (OAuth a sending account) · placement (inbox-placement canary)
smartlead	connect — API key + pick Smartlead mailboxes into the pool (send-only)
domains	list · pause · resume — provisioned OneShot domains
find	watch · drain · import --csv --play · enrich-linkedin · research-prospects · research-queue · research-products · synthesize-angles · score-prospects · backfill-fit-reason · calibrate — --fail-on-empty makes watch --once and drain exit 2 on a run that produced nothing
motion	post-funding concierge demo-no-show competitor-switch hiring-signal podcast-guest discovery-interview free-pilot — each takes --target ; breakup-revive reads the ledger
cadence	advance — poll inbound, fire due steps
direct-mail	list · upload · preview · refresh · approve · send · cancel · skip — individual physical-mail approvals
discover	icp interview-prep · icp synthesize · pmf classify · pmf survey · pmf survey-collect
measure	benchmark — compare this install's command activity with the opt-in telemetry cohort; supports --json
intel	advise · personalize · triage-replies · backfill-intent · weekly-review
handoff	readiness · templatize · first-ae