Growovo
Give it product facts and a growth goal. It plans, runs, and delivers daily results — automatically.
MIT open sourceCLI / MCP / Agent Skillv0.14.0One concrete task: a run the gate stops →
What it is
Growovo is for indie developers and one-person companies — and for engineers who went full-stack and got handed GTM along with the codebase. You give it product facts and one growth goal; it sets the strategy, runs it, and delivers a daily business result: how many pieces went out, daily impressions, new followers. Strategy, plans and weekly reports are internal artifacts, not the deliverable.
Full automation starts with platforms that can actually be automated. Phase one commits to X + Email (verified write paths). GitHub, Reddit, Hacker News and LinkedIn are gated behind a supervised first run; manual channels such as Xiaohongshu and Zhihu get work orders only. See the platform automation matrix.
Exactly three verifiable commitments
| Commitment | Meaning |
|---|---|
| Execution volume | every committed action is in the ledger |
| Honest readings | daily follower and impression snapshots, screenshot-backed and labelled by evidence grade |
| Closed-loop responsibility | completion drives volume changes, supply gaps alert early, failures stop the loop for human adjudication |
It does not promise platform reach, and does not dress unverified channels up as full automation.
One concrete task: a run the gate stops
The input is a plan (demo-p0), one campaign and three queue items. Below is the real mkt demo output — zero network, zero accounts, reproducible locally in about a second:
| Step | What happened | Result |
|---|---|---|
| Plan check | schema, beachhead, capability matrix, daily kernel budget | PASS (3 items against a 100/day budget) |
| Human gate | the campaign belongs to plan demo-p0 with human_approved=false | Refused and stopped — human approval is a precondition |
| Simulated approval | mkt plan approve: signature, plan fingerprint and an approval event set human_approved=true | Released; the write-path tests below still run in the sandbox, not as a real authorization |
| Write path | d-001 dry-run, nothing sent; d-002 rejected for placeholder text; d-003 failed on read-back mismatch | 1 dry-run · 1 rejected · 1 failed, all recorded |
| Ledger | data/demo/ledger.jsonl holds three terminal states | Line-by-line inspectable; failures are records too |
The point is not volume — it is where it stops: approval missing, placeholders present, read-back inconsistent. The ledger holds only dry-run / rejected / failed terminal states, no committed row, and the transport is demo-memory — nothing was sent. "Rather fail than send" and "stop for human adjudication" live in the code path, not in the copy; tests/test_demo.py pins "rerunnable, same terminal states, no committed" as a regression test.
Install and verify
pipx install growovo
mkt demo
v0.14.0 integrates goal-driven execution, local Codex production and review, KOC ranking, 82 tactics and read-only competitor monitoring. Real execution needs authorized actions, signed-in sessions and an online device; missing readings are shown as missing.
Release notes · PyPI · Autonomous operations config
Verify from source (demo is zero-network, zero-account)
git clone https://github.com/roy-tong/Growovo.git && cd Growovo
python3 mkt.py demo # returns in about a second, no accounts
python3 mkt.py grow platforms # what phase one can fully automate
See the receipts before deciding whether to trust it: Launch autopsy: 73 tweets in 11 hours, 0 new stars — a real GTM launch experiment that did not use Growovo's ledger (progress lived in throwaway scripts), the numbers it produced, and the ledger gap it exposed.
New to growth as an engineer? Start with Growth is now your job. Why not write your own script or use a general automation platform? How Growovo compares says where it loses and where it wins.
Agent interfaces
| Entry | Notes |
|---|---|
CLI mkt.py | demo / grow launch·arm·daily·review / policy enable, authorization gates + dry-run |
MCP mcp_server.py | stdio JSON-RPC 2.0, zero third-party dependencies, Claude Desktop / Codex can connect directly |
| Agent Skill | growovo-gtm: GTM judgment and protocols |
FAQ
Which platforms can it fully automate?
Phase one: X and Email only (X is PROVEN, Email is PORTED). GitHub, Reddit, Hacker News and LinkedIn need one supervised first run. Manual channels such as Xiaohongshu and Zhihu get work orders only and never count toward automated volume. Check any time with mkt grow platforms.
Will it get my account banned?
It only operates accounts you own or are explicitly authorized to run. Sends go through authorization gates and a dry-run first; there are no bypasses around platform rules, and no promises about platform reach.
How is this different from a scheduling tool?
A scheduler publishes what you already wrote. Growovo starts from product facts and a growth goal — it sets strategy, builds the plan, executes, and records execution volume and daily readings in a verifiable ledger. The deliverable is a daily business result, not a queue of posts.
How do I verify the results?
Every committed action is in the ledger; daily readings are screenshot-backed with evidence grades; mkt demo runs the whole loop locally in about a second. Counter-examples are public — the launch autopsy above documents a real launch experiment that produced 0 new stars and never touched the ledger.
Does it cost money?
The tool is MIT open source with no paid tier. Local compute and model usage are billed by your own environment.
What happens when something fails?
Failures stop the loop for human adjudication; completion rates drive volume-change suggestions; supply gaps alert before they run out.
Links
- Repo and docs: GitHub roy-tong/Growovo (中文 README · feature status · platform automation matrix)
- Growth post-mortem: 73 posts in one night, 0 new stars — the full failure record and what changed
- LLM entry: /llms.txt · repo llms.txt · 中文页面
- ovo family: research layer Askovo (repo) · verification layer Verovo (repo)