Mantyl turns your repository and coding-agent history into a verified, recipient-ready handover — what exists, how it works, what was decided, what was proven, and what remains unknown.
AI made implementation cheap. It also made understanding expensive: software is easier than ever to create, and harder than ever to trust, transfer or own. That gap has a name — transfer debt.
Decisions, failed approaches and warnings live in discarded transcripts and closed context windows. The code runs — the story of how and why it runs has already disappeared.
Documentation that echoes what an agent said is not evidence. The code says one thing, the transcript says another, and nobody has checked which is true.
The client gets a repository and a deploy key. The next engineer rediscovers architecture, dependencies and unfinished decisions from scratch — and calls you for months.
Every statement in a passport carries its provenance. We don’t generate confidence by hiding ambiguity — we generate confidence by organising evidence.
| Agent said | Repository shows | Mantyl verified | |
|---|---|---|---|
| What it is | Intent, reasoning and remembered outcomes from your coding-agent sessions | Current files, configuration, dependencies and Git history | Commands successfully executed and checks independently completed |
| What it's good for | The why — decisions, trade-offs and abandoned approaches | The what — concrete and current, straight from the source | The proof — strongest evidence, with scope and limitations stated |
| Where it fails | Fallible: may describe code that was never shipped | Incomplete: cannot explain why the design exists | Honest: must record failure and uncertainty, not hide them |
A house buyer doesn’t take the builder’s word for it — they pay a surveyor for independent assurance. Mantyl holds that position for software: the verification layer between an AI-built project and its next owner.
Documentation is not the product. Trust is. The mark is revoked if the evidence behind it can’t be reproduced.
Handover is part of delivery, not administration after it. Mantyl runs where the project lives and produces a meaningful draft on the first run.
Gather the selected agent sessions, Git history and CI output. Local by default — you choose what's included, nothing is uploaded silently.
Index architecture, dependencies, services, environment variables and unfinished work from the repository itself — not from memory.
Execute approved checks: clean build, tests, types, environment. Claims either reproduce, or they get labelled as unverified.
You approve, correct and redact the draft. Sensitive history and inaccurate inferences never publish without a human pass.
A recipient-ready passport: verified facts, honest unknowns, linked evidence — and an acceptance record for both sides.
Narrow integrations, deep output: Claude Code and TypeScript projects first — done properly, before claiming universal compatibility.
Architecture, setup, decisions, risks, unfinished work and verification results — every claim labelled with its source, every section linked to evidence, ending in acceptance rather than optimism.
Verified, repository-backed and agent-only claims are kept apart — and unknowns are findings, not footnotes. Visible uncertainty is safer than manufactured confidence.
Environment, dependencies and run steps verified from a clean machine, not copied from a README that was true three weeks ago.
Key decisions and abandoned approaches recovered from your agent sessions and recorded with dates and links — before they vanish.
The recipient acknowledges what was delivered, verified and left outstanding. You keep a defensible record of the transfer.
Read the whole document: the full example passport for booking-app — all eight sections, including the check that failed.
The recipient is not an afterthought — a passport is only working when the person inheriting the project can act on it.
“I built it quickly — now the delivery needs to look as professional as the product.”
Save handover time, look credible at the moment it matters, and keep a defensible record of what was included, verified and left outstanding.
“Every team hands projects over differently. Clients notice.”
A repeatable handover standard across developers and coding agents — less rediscovery on returning projects, margin protected after launch.
“The app works today — can another engineer own it tomorrow?”
Independent checks, clear unresolved risks and a starting point for the next engineer — before acceptance, payment or contractor exit.
The CLI, the passport format and local analysis cost nothing, ever — no account required. Payment starts where independent verification starts, priced per project like the survey it is.
Free. Forever. No account.
£0 FOREVER
One-off · per project
£19 · £29 · £49 ONE-OFF
For repeat deliveries
£99 / MONTH
Early pricing, proven with alpha partners project by project. passport.json stays open — the format is free so the ecosystem can build on it.
Raw code and transcripts are processed locally by default. Nothing needs to leave your machine to get a draft passport.
Agent history can hold credentials, customer data and private reasoning. You review and strip before anything is shared.
The owner decides what a recipient sees — sensitive history and wrong inferences never go out blindly.
We never hide failed verification to make a passport look healthier. Visible uncertainty is the product working.
Documentation describes software. Agent memory continues it. Observability explains it.Mantyl transfers responsibility for it.
Documentation describes software; Mantyl transfers responsibility for it. The difference is provenance and verification: every claim is labelled as agent-said, repository-shown or independently verified, checks are actually executed, and the output ends in an acceptance workflow — not a longer README. Documentation that merely repeats what an agent said is exactly the problem.
A prompt produces prose from one session's memory. Mantyl ingests history across sessions, inspects the repository directly, runs deterministic checks, detects contradictions between what was said and what exists, and forces a human review-and-redact pass before anything is shared. The value is the evidence chain, not the text.
The CLI, local analysis, passport.json and the HTML report are free, forever — adoption should cost nothing. You pay when independent trust enters the picture: the Mantyl Verified accreditation, checks executed on clean infrastructure, hosted reports with an audit trail, and CI re-verification. Documentation is free; trust is the product.
Nothing, by default. The CLI is local-first: analysis runs where the project lives, you review and redact the draft, and only the passport you explicitly approve is published. Transcripts are treated as sensitive material — they can contain credentials, customer data and private reasoning.
That's why it's priced per transfer event, not as another subscription you have to justify. Freelancers generate a passport at delivery; agencies standardise it across client work; recipients can commission an incoming audit before accepting a codebase.
Claude Code and TypeScript projects first — narrow integrations, deep output. It is better to support one coding agent and one common stack properly than to claim universal compatibility with shallow analysis. More follow once the first is excellent.
We’re onboarding AI-first freelancers and agencies with a client handover in the next 30 days. The first passports are concierge — built with you, on a real project, so the output is judged on a real transfer.
Local CLI · Claude Code + TypeScript first · Nothing published without approval