WsTree
Autonomous development operation

The dev that never sleeps.

WsTree is an operation of autonomous agents that runs your development cycle for you — and keeps you in the loop via Telegram. You decide. The tree executes.

24/7 in operation100% audited1 tap to approve
wstree — tuesday, 3 am
The system

An engineering operation,
not a chatbot.

WsTree puts AI agents to work like a real team: it schedules, executes, pauses, resumes and audits every task — with Telegram as your control room.

🌳

Real autonomy

Scheduled workflows run on their own: bug triage, PR review, QA, features. Every run has scope, tools and limits defined by you.

🧑‍✈️

You in the loop

Agent stuck on a decision? The run pauses, the question lands on your Telegram, you answer from your phone and the work continues exactly where it stopped. Autonomy without losing control.

🛡️

Safe foundation

Every agent action goes through a deterministic security layer before it executes. Restricted scope, secret scanning, explicit confirmation for privileged actions and auditing of everything.

Before × After

Traditional dev × 24/7 operation

It's not about replacing you — it's about promoting you. You become the tech lead of an operation that never stops; the grunt work runs on its own and the decisions remain yours.

Traditional flow
With WsTree
Working hours
8 hours a day, 5 days a week — when there are no meetings.
Continuous operation, 24/7. Late nights, weekends and holidays included.
Bug reported at 11 pm
Goes into the queue. Someone looks tomorrow… maybe.
/bughunter reproduces the bug, takes screenshots and proposes the fix in an isolated environment.
Context
Every interruption costs 20 minutes to recover the thread.
Persistent sessions per workspace — context is never lost between messages.
Parallelism
One task at a time, a single brain.
Background jobs with a queue per workspace, multiple AI accounts and several simultaneous runs.
Critical actions
Friday deploy? Ceremony, fear and crossed fingers.
Exact scope on screen + explicit 1-tap confirmation — the decision is always yours, never the model's.
Accountability
"What did you do yesterday?" — depends on memory.
Every action, token and cost in a complete, queryable audit trail.
In practice

How it changes your life

Real scenarios from operating a tree: the system works, you steer.

☀️

Wake up to ready PRs

Overnight workflows triage cards, hunt bugs and open PRs for your review. Your morning coffee comes with work delivered.

/pr list
📱

Command from anywhere

Fire off a bug hunt from the grocery line. The agent reproduces, screenshots and proposes the fix — you just approve.

/bughunter card-231

Approve with one tap

Professional PR review becomes an immutable draft: what you read is what gets posted. Approve, merge and retarget right from the chat.

/pr approve 42
📎

Send a screenshot, get a fix

Attach a screenshot, PDF or CSV to the conversation. The context enters the next run automatically — with real file-type validation.

send photo + caption
🔀

Many clients, one bot

Isolated workspaces per project, each with its own session, tools and repositories. Multiple AI accounts with automatic rotation.

/ws client-x
📊

Cost under control

Every token becomes an audit event: cost per day, per model, per workflow — always at hand, right from the chat.

/digest
Human-in-the-loop

Autonomy isn't absence.
It's knowing when to call you.

01

The agent finds a decision that's yours

Every workflow carries a protocol: facing a business choice, the agent stops and asks — instead of guessing.

02

The run pauses and the question hits your Telegram

The system saves the session, pauses the work and notifies you. Nothing moves without you — and nothing is lost while waiting.

03

You answer, the work continues

Your answer resumes the exact session. The agent picks up with full context and delivers — hours later, if needed.

🌳

WsTree

bot · online

bug-triage · run #483 · paused

❓ The fix requires removing the legacy_checkout feature flag. It looks dead since v2. Can I remove it?

Go ahead — that flag really died in v2. 👍

▶️ Resuming the session…

✓ done — flag removed, tests green, PR #87 open for your review.

/pr approve 87
Security

Built to run alone
without costing your sleep.

Autonomy is only acceptable with layered defense. WsTree's rule: whatever is critical is decided by deterministic code — never by the model.

🔒

Locked control room

The bot answers one single authorized chat. Anyone else is ignored — and logged.

🗺️

Filesystem boundary

The system only sees and touches the directories you authorized — enforced in independent layers.

🚦

Deterministic guard

Every agent action is validated by deterministic code before it runs. Destructive operations are blocked — the model can't override it.

🔑

Secret scanning

Pushes and deploys are blocked if the scan finds a real key, a hardcoded credential or a forbidden file — even inside the build bundle.

☝️

Conscious confirmation

Deploys, merges and privileged writes require the exact scope on screen plus your explicit confirmation tap. The decision is always yours.

🧾

Full audit trail

Every action and every token becomes an append-only audit event. Fail-open: auditing never breaks a run.