Private AI assistant & operations manager
A private assistant that runs on your own laptop, remembers your notes, does the recurring jobs, and hands the hard cases to Claude. We run one. We build yours.
you make a pong game for the browser route longer code -> coder 7b (loaded, 4.7 GB) lil wrote games/pong.html and opened it. 2 paddles, ball speed-up, first to 7. you summarize my investing notes route notes -> chat 3b (from your 27 saved notes, nothing else) you the ball clips through the paddle at high speed route bug fix -> /claude (file + message copied, paste into Claude) 08:00 job digest: 31 boards checked, 6 new revenue-ops roles -> tasks/digests/ 08:05 morning summary written. 5 sites up. itch.io rate-limited (not down).
We built our own first. It runs on a MacBook with open models, no API bill and no data leaving the machine. It routes each request to the smallest model that does it well, keeps notes it answers from, runs the recurring jobs (site checks every 30 minutes, a morning job digest from 30 boards, a daily summary) and copies the hard cases to Claude with one key. On top of it, a Claude-based operations manager runs six ventures from one board, keeps long jobs alive overnight and checks the live result before reporting. We build the same two layers for your business, on your tools, with your rules.
What you get
Every engagement is scoped in writing before work starts. These are the usual deliverables; take the ones you need.
Local assistantOllama, 3B to 7B models
Open models installed on your Mac or PC, a chat window, a command-line ask, and notes it remembers and answers from. Works offline. Nothing is sent anywhere unless you say so.
IncludedModel routingRouting rules plus test set
Each request goes to the smallest model that does it well: a chat model for text, a small coder for scripts, a larger coder for apps and pages, and a one-key handoff to Claude for bug fixes and exact-format work. Tested on a sample set before handover.
IncludedRecurring jobsScheduler plus digests folder
The checks and digests you do by hand, run on a schedule on the machine: uptime checks, job or lead digests, a morning summary. Public pages and APIs only; no logins, posting or sending without you.
IncludedOperations board and rulesBoard plus rules file
One live board every AI session reads first, and a standing-rules file it never forgets: your voice, approvals, what never gets sent or spent without a person.
IncludedAgents that survive the nightCheckpoint, relaunch, quality gate
Long jobs run as background agents that save progress every 20 minutes and relaunch if they die. Done means verified by someone who did not do the work.
IncludedHandover and trainingLive, with recordings
A walkthrough, the rules and routing files, and a month of tuning as it meets your real week.
IncludedWhy this is a known quantity
The founder's in-house record on this exact work, before Known Quantity existed.
$0 per month
to run the local assistant: open models on a laptop, no API bill, no subscription, no data leaving the machine
See the results26 of 28
routing tests passed on the first run; the two misses were fixed and re-tested before we called it ready
See the results6 ventures, 1 board
run by the operations layer, overnight with the laptop closed, verified before reported
See the resultsHow it goes
- Step 1 of 3
Scope it in writing
A few emails to pin down the problem, the data and what "done" looks like. You get a written scope and a price before anything starts.
The full process - Step 2 of 3
Build on your real data
Days, not months, for most projects. We say which parts were automated and which were checked by hand.
How we use automation - Step 3 of 3
Hand it over, documented
Files, logic and a walkthrough. Nothing depends on us to keep running.
What that looks like
Questions people ask
What does it run on?
Your own Mac or PC. A 16 GB MacBook runs the 3B chat model and the small coder together and loads the 7B coder on demand. Models unload after five minutes idle so the machine stays usable. No GPU rental, no server, no monthly bill.
Is it as good as Claude or ChatGPT?
No, and it does not pretend to be. Small local models lose facts and hallucinate on hard questions, so the assistant answers facts about you only from your notes, and sends bug fixes and exact-format work to Claude with one key. You get privacy and zero cost for the 80% of requests that do not need a frontier model, and a clean handoff for the rest.
What does the operations manager add?
The local assistant is yours alone. The operations layer runs on Claude and holds the board, the standing rules, the background agents and the quality gate, so work continues while you sleep and nothing is reported done without evidence. Most clients start local and add the operations layer when they have more than one thing to run.
What does it cost?
The local assistant build is a fixed $3,500: install, routing, notes, recurring jobs, handoff, handover and a month of tuning. Local assistant plus the operations manager is $7,500. The playbook we use for the operations layer, with the board template and rule files, is sold on its own for $49.
What about our data?
The local assistant keeps everything on the machine. Anything that goes to Claude goes by hand, on tiers that do not train on your data, and the rules file says what may never be sent or spent without a person. Those rules are written before it touches anything.
Related work
AI consulting & automation
Where AI actually pays off in your finance and revenue operations, built and proven in your numbers, not a slide deck.
Read moreSales compensation & quotas
Commission plans that pay the right amount, on time, with an audit trail reps can read.
Read moreReporting & dashboards
Reports that refresh on schedule and land in the right inbox, so nobody rebuilds the Monday file by hand.
Read moreReady to scope ai assistant work?
Send a few lines about the problem and the tools involved. You will have questions or a quote within one business day.