clembot.wanessalabs.com
An AI coworker you can audit.
Clembot drafts, reviews, schedules and logs marketing work — and writes down everything it did while doing it. This site is built from that record. Not a case study about the record: the record.
By the numbers
the arc
13 weeks planned, 15 weeks in
Started 2026-04-24. Planned through 2026-07-23. Currently week 15. Which is 2 weeks past the planned end. The plan is left as it was written rather than restated to match what happened — a 13-week plan that took 15 weeks is a more useful thing to publish than a 15-week plan that finished exactly on time.
- 01
Foundation
- marketing-bootstrap scaffold complete
- clembot-vault structured with project template
- clembot/sabbatical and clembot/linkedin scaffolded
- distributor v1 specced (Instagram adaptor)
+ 1 goal withheld — named client work
- 02
Build
- Notion pages fully structured and populated for each active project
- Dashboard reading live vault data (not mock)
+ 4 goals withheld — named client work
- 03
Launch
- First paying Clembot client signed (upsell from Slack community)
- Clembot install package documented and repeatable (onboarding playbook)
- Public launch post — "here's what I built and what it cost"
- Pricing confirmed and live: $100/month retainer (client pays own Claude Code usage)
Phase status comes from the project config as written. Where the config and the handoff notes disagree about progress, the config is shown unedited rather than reconciled. [VERIFY]
how it works
Six rules doing most of the work
A human approves everything that leaves the system
No timer publishes. Every outbound post passes a person first, and the switch that would change that ships defaulted off.
02One writer per table
A single command-line tool owns the content lifecycle. The publishing tool and the client-facing workspace are projections of it, never sources of truth.
03An unverified number gets a label, not a guess
If a figure isn't in the config, it ships marked [VERIFY]. A plausible number is worse than a missing one, because nobody checks it.
04Redact by allowlist, and fail closed
Everything published here is built field-by-field from a literal, scanned before it is written, and the scanner aborts the whole export on a single hit.
05The log is the product
Every agent action writes an evidence record. The records are hash-chained, so editing history is detectable rather than merely discouraged.
06Code and client data live in different repositories
The engine is one repo. Every install's data is a separate vault the engine reads through an environment variable. They are never mixed.
output
What actually shipped
- Log of Logs ↗Shipped
- Clembot-dictate (voice transcriber)Shipped
- Sudoku Book Maker ↗Production
- Vault KanbanProduction (POW verified)
- Wanessa Labs Games ↗Production
- wanessalabs.com ↗Production
- Learn to DrawPlanned
- Marketing Bootstrap (Clembot)Active — buzz-hardening merged (PR #5): signed/tamper-evident/searchable audit substrate (evidence hash-chain + Ed25519 signing + verifiable approvals + FTS5 ev
Want one of these?
Clembot installs are single-tenant: your own vault, your own configuration, your own approval gate. The waitlist is open and small on purpose.
Join the waitlist