Donezo logoDonezo Download

Staying in control

See what needs you, read what agents did, and stop them when you need to.

Home inbox

The Home page. Needs you lists a message waiting for approval with Deny and Approve buttons, and failed sub-agents with an Open in Team button. Below are workspace cards. On the right, Working now lists running agents and Today counts runs, verified and failed work.
Home, across every workspace.

Needs you collects two kinds of item from all workspaces. The number in the sidebar counts them.

Other messages from agents do not appear here. A director tells you things in its own terminal, so keep an eye on it.

The side column shows Working now, which agents are running, and Today, the counts of runs, verified work and failed work. Below the inbox, each workspace card says needs you, how many are running, or idle.

Approvals

A message needs approval when its sender asks for it: an agent marks a message as sensitive, or you tick Requires approval under More options in Send a message. It then waits. It does not run until someone approves it.

You can answer in several places:

A receiving card wears a needs you badge in the meantime. A denied message stays in Activity and can be run later with ↻ Re-run.

Activity

The Activity tab of a workspace is one list of every message between agents and every run. Items still in motion are under Now, the rest under Earlier.

The Activity tab. Filter buttons sit above a timeline of runs and messages, each with its time, who it involves, and its status. A run is open on the right, showing its transcript step by step.
Click a run to read its transcript. A run still going keeps updating.

Some automatic messages Donezo sends agents are currently written in Vietnamese, such as the [BOOTSTRAP SKILL …] one on an agent's first run. You do not need to act on them.

Guard rails

All of these run without you doing anything, unless a default says "off". Variables go in your .env file.

GuardWhat it doesSet with
Ping-pong capTwo agents may exchange 2 messages per task; further sends are refused, and the agent is told to report to you. The count resets when you give either one new work, or after 60 minutes of silence. See Signal budget in the agent's panel.ORCH_PAIR_SIGNAL_CAP, ORCH_PAIR_SIGNAL_WINDOW_MIN
One prompt per agentQueues messages, so a conversation is never written by two runs at once.Always on
Runs at onceAt most 3 background runs at the same time across the app.ORCH_MAX_CONCURRENT
Daily run limitOff by default. Once reached, messages to that agent are blocked. Allow + in its panel adds 10 more runs for today and puts those messages back in the queue.ORCH_MAX_RUNS_PER_DAY, ORCH_DAILY_ALLOW_STEP
Lifetime limitsOff by default. A total number of runs or tokens per agent; beyond it, messages are blocked.ORCH_MAX_RUNS_PER_SESSION, ORCH_SESSION_TOKEN_BUDGET
Fix attemptsFailed verifications before a sub-agent's work is marked failed: 3.ORCH_SUBTASK_MAX_ATTEMPTS
Who may message whomOptional allow-list per workspace.Team rules
Your machine onlyListens on 127.0.0.1 and refuses requests from other websites. Read the security warning before changing this.ORCH_HOST

Kill switch: how to stop things

ControlWhereWhat happens
Stop buttonLeft strip of a card, only while a job is running. Its tooltip reads Kill the running job (stops a runaway).After you confirm, the run is killed, marked failed, and not retried.
Pause / ResumeLeft strip of a cardWhile paused, messages to that agent wait. Resume runs them.
Suspend / ActivateTop right of a workspaceWhile suspended, messages in that workspace wait and no new agents can be created. A job already running is not killed; use the stop button for that. Activate runs what waited.
CancelA card on the Team tabStops one sub-agent's work and tells the director.