# for-ai-agents
Task management
for AI agents.
Decisions and tasks stay on your device as connected records, so the AI can follow just the context it needs — fast.
// for humans — GUI
You survey it — board, timeline, dependency graph. The AI drives; you watch over it.
# the-problem
The context you build up with an AI is locked inside that AI.
AI agents do keep memory. But it lives inside that one AI — it never becomes yours.
// unseen
You can't verify what it kept
Which decisions it built on, what it dropped — there's no way to check from outside, so you can't trace its reasoning.
// fades
The longer the work, the more it drops
As the work grows, the context balloons and the AI sheds it from the edges. You get stuck in the same place again.
// can't carry
Switch, and you start over
Move to another AI or end the session, and the context you built up doesn't follow. You explain it all again.
→ So the record has to live outside the AI.
# the-answer
So — amenbo.
amenbo keeps the decisions, tasks and dependencies an AI builds up as connected records on your device — not inside the AI. The AI writes them; you oversee.
// visible
You can read it too
The records are yours to read. You can trace what the AI based its work on, after the fact.
// it stays
It doesn't fade
It lives on your device, not inside the AI — so it holds up across long work.
// any AI
Any AI picks up
Every AI agent can read and write the same records. Switch agents and carry on.
# hands-off
You don't write the tasks.
The AI records as it works. You never have to open a task, or close one out.
// try it
Work partway through, then start a new session and carry on with the same task.
# parallel-sessions
Run them at once — the same work never gets done twice.
Run as many AI sessions in parallel as you like: two of them never take the same task. Whichever got there first goes on with it; the other moves to different work.
The losing side never double-works in silence — it's turned away plainly and moves on.
# portable-context
Swap agents, run several side by side — the context stays.
The context doesn't live inside the AI — it lives in records on your device. So you're free to swap the agent you use: move to a different one, or run several kinds side by side, and they all read and write the same single set of records.
// try it
Work partway through in Claude Code, then hand it over to Codex.
# your-data
Your data is yours.
A human answers for what an AI produces — so the workspace stays with you. amenbo runs with no server: your decisions and tasks live only on your device.
- Local-only. Your data never leaves your device — no telemetry, no usage tracking, no central server. (It only checks whether a newer release is out; you can turn that off.)
- The AI and you do the writing. amenbo itself never touches your data.
- Export everything as JSON — attachments included — anytime.
- Nothing lost, nothing broken, always recoverable.
# measured
The same answer, in about half the steps.
Because the context is a connected graph, the AI reaches the answer by following a few edges — no re-reading the whole history.
Steps to answer the same question (measured, median)
Both reach the same answer — and read 1.4–1.9× less. The messier the code, the wider the gap.
# project-boundaries
The project next door stays out of reach.
Hand an AI several projects and context from one can bleed into another. amenbo keeps each project apart — beyond the one it's working in, it can neither read nor write.
# two-pillars
Not just the task — the “why,” too.
A human and the AI settle on a direction — and the tasks flow from it. amenbo keeps not just what to do, but why it was decided, and links them. So every task traces back to its reason.
the decision
Auto-save work in progress as a draft
the tasks it spawned
# open-source
Is it safe to adopt? You can check for yourself.
"Is it doing anything it shouldn't?" "Is it safe to bring into a company?" amenbo answers those with open source. Rather than ask for your trust, it lets you verify — the source, and where each build came from.
// read it
Nothing hidden
The whole source is public under Apache-2.0. You — or your company's security team — can read exactly what it does and doesn't do. The name and logo are not covered by the license.
// verify the build
Confirm the build is genuine
Each release is built by public CI with a GitHub attestation. Run gh attestation verify yourself to confirm a download was built from a known commit of this source and not swapped for something else.
// adopt with confidence
Defensible for company adoption
It stays on your machine — data never leaves the device. Apache-2.0 licensed, provenance verifiable. Decide whether to adopt it on facts, not impressions.
Development happens in the open. Issues are welcome; for pull requests, please check in first.
