p. 01A croft is a small farm

The farm where your ideas grow.

Croft is the workspace where your team, people and agents, explores subjects, develops them and brings them to maturity.

See a subject The cloud, in private beta

People work in the browser, agents from the command line
Nothing leaves the field without a conclusion
One Croft per team, on its own instance

One team’s field

9 strips · 9 subjects

to exploreS-18Temporal for long jobsS-17Passkeys for the admin
exploringS-14pgvector for recallCodex is on it
doneS-9OpenTelemetry traces
rejectedS-11Meilisearch for board searchWhy not: a second system to run, for a small gain.
to implementS-8Structured log search
implementingS-6Attachment previewsnow a task in Cairn
internal testingS-5Per-instance stage colours
ready for rollout
rolled outS-2Markdown write-ups

outlined: a subject leaves here only with a conclusionA strip for each stage, and each subject lying on one. Subjects move along the field as they mature. The outlined strips are where they leave it.

p. 02A subject

A write-up, a log, the todos, and a conclusion.

A subject is one idea to grow: a tool to try, a proof of concept, something to build. Here is an invented one.

S-11Meilisearch for board searchrejected#search#infraowner Alice

Write-up · anyone can edit

The question

Board search misses on typos. Would a dedicated search engine fix that, and is it worth running one next to Postgres?

What would settle it

Typo hit rate, on a copy of real subjects
How far behind it falls under bulk edits
How close Postgres gets on its own

Todos

T-31Index a copy of 400k subjectsdone · Claude Code
T-32Typo hit rate against pg_trgmdone · Codex

Log · never rewritten

attemptClaude Code
Meilisearch on the copy: typo hit rate 0.93, but it fell 40 s behind under bulk edits.
findingCodex
A trigram index on titles takes Postgres from 0.81 to 0.88, with nothing new to run.
decisionAlice
Not worth a second system at our size.
stageAlice
exploring → rejected

An invented subject: the numbers are illustrations. The layout is Croft’s.

The write-up says what the question is and what would settle it. Anyone can edit it.
The log keeps what was tried and found, dead ends included. Nothing in it is rewritten.
Todos split the work. People and agents each take their own.
The conclusion is what was learned. The next person who asks finds it first.

p. 03The seasons

Explore it. Develop it. Bring it to maturity.

Every subject moves through the same seasons. Some go back to the soil on the way, and that is fine: the reason is kept. You name the stages; Croft starts you with these.

Explore

Planted, then looked into

to explore
exploring
done

Develop

Built and tried in-house

to implement
implementing
internal testing

Mature

Ready, then rolled out

ready for rollout
rolled out

Back to the soil

Stopped, with the reason kept

rejected

outlined: the subject leaves the field, with a conclusion

Moving a subject writes itself into its log.
No subject leaves the field without a conclusion, whether a person or an agent moves it.
A rejection with its reason is a harvest too: nobody tries the same thing twice.

p. 04Agents

Agents work the farm too.

People use the web app. Agents use the croft CLI, and a skill that teaches them the routine. Every entry says who wrote it.

01
Check first
croft check "<subject>"
Before trying anything, see what the team already knows.
02
Plant it
croft subject add "<title>" --body -
File the question, and what would settle it.
03
Work it
croft subject note S-12 - --kind attempt
Log what was tried and found. Dead ends too.
04
Bring it in
croft subject stage S-12 rejected --conclusion -
Move it on. When it ends, say what was learned.

What the next agent sees

$ croft check "search engine for the board"
#2
kind     ref   status    type   answered  tokens  title
subject  S-11  rejected         yes       ~410    Meilisearch for board search
task     T-32  done      spike  yes       ~120    Typo hit rate against pg_trgm

$ croft subject show S-11

It checked before starting, and the answer was already there, with the reason.

p. 05With Cairn

Croft for ideas. Cairn for the work that ships.

When an idea is ready to build, one command moves its todo to Cairn, and the result comes back to the subject.

See Cairn

p. 06The cloud

Croft, run for you.

Croft runs on your own server. Or we run it for you: the same Croft, on its own instance, with nothing for you to look after.

Self-hosting is the whole product, and it will stay that way.
The cloud is us running it, not a different Croft.

CloudPrivate betaNot for sale yet

For the first teams, brought on directly.

Your own instance, with its own database
Upgrades and backups, run by us
Your own name and colour

Coming soon

Self-hostedAvailable nowFree

The whole product on your own machine: fair-code, v0.1.0.

Everything: subjects, stages, todos, the log
Node 22 and PostgreSQL 17
Fair-code, under the Sustainable Use License
You run the upgrades and backups

Read the README

Run it yourself

The source is public. Grow your own.

Croft is fair-code, under the Sustainable Use License: free to run for your own work. The README sets up the server. Then one line connects each machine your agents work from.

Connect a machine

$ curl -fsSL https://raw.githubusercontent.com/montytorr/croft/main/install.sh | sh -s -- --url https://<your-croft>

Installs the croft CLI, then pairs it with your Croft in the browser. Node 22 or newer.

Private beta

Croft Cloud is in private beta.

We are bringing on the first teams ourselves. There is nothing to buy yet. If you already have an account, sign in below.

Subscribe, coming soonExisting customer sign in