Charter
An AI co-founder · for one founder, not the next 10

Idea to launched MVP,
with judgment that compounds.

Charter runs a portfolio of small software bets — scored against your real constraints and sharpened by every bet you’ve already shipped or dropped.

Get early access →How it works ↓
charter.app · Studio

Good morning, Jess.

tue · jun 02 · 09:14 ET
4
in flight
12
in idea inbox
2
need a decision
DECISION DUESwitchback hit its 30-day at-risk window. Drop, extend 14 days, or pivot scope?
Bonsai Compose
PR Reviewer agent prototype lands tomorrow.
ON TRACK
DAY 12 / 90
Tideline
Concept doc complete, advancing to Design.
ON TRACK
DAY 5 / 60
Plot.garden
$140 MRR, week 2 — eight new paying signups.
SCALE
DAY 47 / 90
Switchback
At risk: ad clicks down 40% week-over-week.
AT RISK
DAY 31 / 90
If this sounds familiar

You know how to build. That was never the problem.

The hard part is everything around the code — picking the right bet, knowing if it’s working while it’s still small, and remembering what killed the last one when you start the next.

“I have eight ideas in my notes. Three look promising. I can’t tell which one is worth the next ninety days.”
The triage problem
“I’ve been on this for two months. Three signups a week. Is this working? Should I keep going?”
The no-signal problem
“Last bet died for a reason I should have seen at week three. I’m tired of learning the same lesson twice.”
The memory problem

Charter is the layer around the build— the part that picks what to take on, watches it honestly while it’s live, and writes down what you should have known. So the next bet starts smarter.

Generic AI app builders will write you code. They’re good at it. They’re also one-size-fits-all — your 12 hrs/week side project and a VC-backed team get the same suggestions.

Charter is not that. Charter is the layer around the code generator — deciding what to build, whether anyone wants it, how to reach them, and when to walk away.

It reads your real constraints. It remembers your past bets. It tells you hard truths kindly. The code still gets written by Claude Code at the end. The judgment is the product.

It’s built for one kind of person: the solo founder shipping small software bets, who can run Claude Code or pair with someone who can — not a team, not a no-code creator.

Generic AI app builder
AI
Build an app with one prompt.
Describe what you want — we’ll build it.
Generate
Same suggestion for every founder. No memory of the last 19 bets.
Charter inbox · this idea, this founder
71convictioncharter:weekly · this morning
“An invoice-amount sanity check for solo consultants — flag missed late fees before sending.”
FIT WITH OPERATOR
Postgres + TS are native here. CLI is fast for you to build, no design or cold-call required. 12 hrs/week is tight against your $3K MRR target — recommend trimming MVP scope to invoice-flagging only.
Read from your strategy doc. Calibrated against your 19 past bets.
Same idea. Different judgment.
Two things make Charter different

Neither of them is the code.

The moat compounds

Bet #20 isn’t scored like bet #1.

This is the one Claude alone can’t do. Every project ends with a postmortem that distills one rule for next time. That rule joins the library and feeds the conviction scorer on every future idea — so the scorer you have after 30 bets is a different scorer entirely. The judgment is the thing that accrues.

BeaconDroppedmonth 2 · phase 4 launch
Thesis
Push-notification digest of your unread Substacks.
Missed
Three free Pocket-style alternatives shipped same quarter we did.
Signal
We never tracked competitive launches per category.
Rule for next time“If a category has 3+ free alternatives launching the same quarter, discount conviction by 20 — and require paid upfront.”
→ joins every future score’s context.
Personalized at the SQL level

Charter knows you — concretely.

Hours per week. Monthly budget. Skills. Gaps. What you’ll build. What you won’t. Every agent reads this profile on every call, so the same idea scores 71 for one founder and 50 for another. On its own that’s table stakes — it’s what the loop above compounds on top of.

Hours / week · Monthly budget
18 hours · $240/mo (Anthropic + Vercel + ads)
Skills (and gaps)
+Native Postgres, TS, devops
+Comfortable writing dev-targeted copy
−Weak at illustration / brand design
−Allergic to phone calls
Headline goal · Specific target
Replace day-job income · $4,000 MRR by month 9
“GTM never suggests cold sales — it reads this line first.”
strategy.profile · updated 4 days ago
The loop

Six phases, one founder, calm by default.

Every promoted idea moves through these phases. At any gate, you can drop the bet or ship the project. Either way, the Postmortem agent writes a rule. The rule joins the library. The next idea is scored against it.

A worked exampleBonsai Composescored 71 at promote · followed through every phase below
Phase 1 · Ideate

Talk it through. One question at a time.

Charter pushes back on the weakest assumption first, one question at a time, until you have a falsifiable bet thesis — written in your own words.

ships in v1
“What’s the cheapest 48-hour test that would tell you nobody wants this?”
Show HN with a CLI that flags missed late fees from a Postgres dump.
“Good. What would kill the bet? Be specific — a number.”
Fewer than 50 stars or 5 paid inquiries in two weeks.
“Logged. Bet thesis ready. Promote when you’re ready.”
CONCEPT DOC · DRAFT
PROBLEM
Solo consultants under-bill by missing late-fee triggers.
AUDIENCE
Independent consultants invoicing monthly via Stripe.
VALUE PROP
Recover ~3% of yearly revenue in 5 mins / month.
MVP SCOPE
CLI · Postgres dump → flagged invoices.
SUCCESS METRIC
50+ HN stars / 5+ paid inquiries · 2 weeks.
BET THESIS
ready
A co-pilot, not an oracle

Calibrated. Not certain.

Charter’s conviction scores carry confidence levels. Its diagnostic notes propose the cheapest next move that would clarify the picture — not a final verdict.

The Ideation agent says “I don’t know yet” rather than guessing. The Conviction scorer flags when it’s working from profile alone, because no lessons have been logged yet.

Because the alternative — overpromising, then being wrong — kills trust on the second wrong call. Charter is honest about what AI can and cannot know about your bet.

58
Confidence · medium-low
manual · 12m ago
“AI code review built specifically for solo devs — calibrated against your own PR history, not the median dev team.”
FIT WITH OPERATOR
Strong skill fit: Postgres and TS are native here, CLI is fast to build, no design or cold-call required. However, 12 hrs/week is tight against the $3K MRR target — solo devs convert slowly through the funnel.
BIGGEST RISK
Willingness to pay. Solo devs hit this pain infrequently, live with it when they do.
CHEAPEST TEST
Post a minimal open-source version on Hacker News “Show HN”. If 200+ stars and 10–100 people ask about a paid tier within two weeks, demand is real enough to price.
LESSONS APPLIED
baseline 72final 58−14 from your lessons
−14“If a category has 3+ free alternatives launching the same quarter, discount conviction.”from Beacon · dropped month 2
Personalized rationalereads the strategy doc, not the median founder
No certaintysays “I don’t know yet” when it doesn’t
For the founders thinking it

Like dropshipping. Honest about how it’s not.

The portfolio-of-small-bets shape is borrowed from dropshipping. The numbers underneath are not. Worth being plain about both.

What survives the analogy

  • +Run a portfolio of small bets at once
  • +Set falsifiable conditions up front
  • +Kill fast when the signal says no
  • +Treat drops as data, not failure

What doesn’t

  • −Software MVPs convert at 1–3%, not 5–15%
  • −A brand new domain has no trust
  • −The iteration cycle is weeks per bet, not days
  • −Diagnosis is probabilistic, not deterministic
So Charter borrows the shape of the loop. The math underneath is yours.
Closed beta · waitlist open

Early access. One seat at a time.

We’re letting a small group of solo founders through the door while we sharpen the loop. Drop your email below — when a seat opens that fits your profile, we reach out.

stage 1 of 2 · takes 30 seconds total

Once you’re in, you’ll pay your own Anthropic API key, your own hosting, your own ad spend. Charter never marks those up. The Charter subscription itself starts at $39/mo when public; beta seats are free.

Public repo · v0 in progress

Built in the open.

Charter is built dogfood-first. Every line of code, every prompt, every architectural decision lives on GitHub — published while we build, not after.

github.com/OceanAtlasKingdom/charterRead the vision doc →
OceanAtlasKingdom/charter · mainlast commit 14 min ago — “close Netlify path leaks — autonomous chain now bi-target”
▾agents 2h
›conviction.v2.md 2d
›ideation.v2.md 3d
›design_audit.v1.md 14m
›audit.v1.md 1d
›postmortem.v1.md 6d
▾app 1h
▸(shell) 1h
›page.tsx 9h
›layout.tsx 2d
▾supabase/migrations 3h
›0023_design_audit.sql 3h
›Charter-Spec.md 1d
›Architecture.md 2w
›README.md 3d
Plain answers

The questions that come up.

Is this just another AI app builder?

No. Charter writes the specClaude Code runs. We don’t compete with the code generator — we sit upstream and downstream of it. Upstream: ideation, conviction scoring, design brief, build package. Downstream: audit, launch, monitor, postmortem. The middle slice — actually writing the code — is the commodity. We hand it off.

Why not just use Notion + Claude, or YC’s Startup School?

You can — and for one bet, you probably should. Those give you the framework: a canvas, a checklist, a postmortem template. What they don’t do is remember. The founder friend who recalls exactly what killed your last three bets and stops you repeating it — that’s the function Charter automates. A template starts blank every time; Charter’s scorer walks in already carrying your scar tissue. That’s the whole bet, and it only pays off across many bets — which is also why it’s honest to say it’s most valuable after your fifth, not your first.

Why personalized scoring?

Because the same idea is a different bet for different founders. A wedding photographer with 8 hrs/week and a backend engineer with $5k/mo budget aren’t betting on the same thing when they say “build me an AI tool.” The honest score reads the strategy doc first, and recalibrates from your own past outcomes — not generic founder advice.

Can I bring my team?

Not yet. Charter is for one founder. Teams may come in v2; right now solo is the entire product surface, and most of the calibration logic only makes sense against a single operator profile.

What does it cost?

Free during closed beta. $39/mo flat when public — a tool subscription, not a freemium tier. Paying unlocks the whole product: no feature gates, no per-bet or per-idea caps. You pay your own Anthropic API key and hosting on top; we never mark those up. The only real ceiling is your own API spend, which you watch directly in the app.

Does it work for non-technical founders?

Yes, for the thinking half. The Ideate, Design, Launch and Monitor phases don’t require code skills. You’ll still need to run the build package through Claude Code somewhere — which usually means hiring a developer or pairing with one. We’re not pretending Charter is a no-code tool.