agenticcodingresearchopsdocumentedunattended capable

Operational brief

Agent Zero

A self-hosted Linux workbench with chat, browser, desktop and file/code tools; optional scheduled work still needs an operator watching the results.

No product visual on file

Choose it when

You want coding and research in one Docker/Web UI workspace, with project memory and an optional bridge to your host files.

Skip it when

You need a managed service, an independently verified isolation guarantee, or a mandatory approval gate before every action.

01 / Classification

Autonomy evidence

This is our classification, not a vendor badge. The record says what the product normally does. The gaps stay visible.

How we classify it

Tool use + multi-step plans, still human-gated per session.

Runs unattended

Yes — that capability is documented in this brief.

Trigger

Normal work starts with a user chat prompt. The optional scheduler runs cron schedules, planned dates or manual tasks; unattended capability isn't the default chat experience.

State

Tasks create or load persistent dedicated/shared chat context. Project-scoped runs load project instructions, secrets and memory; this is context separation, not a claim of separate containers.

Tools

Scheduled runs execute with available tools. The runtime exposes code/file work, desktop apps and browser writes; host file access and remote code execution require separate connector grants.

Human control

The user supplies error handling and checks logs and Last Result regularly. Tasks can be stopped mid-run or disabled. Notifications require configuration; automatic recovery and guaranteed alerts aren't established.

02 / Operations

Control reality

Useful agents eventually touch something real. This is the operational picture we have, not a promise hidden in a sales page.

Where it runs

Agent Zero runs in Docker on infrastructure you operate. Launcher manages setup/instances; A0 CLI connects to an existing instance rather than running a second agent.

Data boundary

Self-hosting the runtime doesn't make inference local. Choose local or cloud model access. The host bridge reaches outside Docker; host-browser debugging grants full session control, while page text/screenshots follow local-only, warn-on-cloud or allow settings.

Identity & permissions

You configure UI login/password and provider credentials; project secrets hold task credentials. Set UI authentication before opening a tunnel. This review doesn't establish enterprise SSO or a universal per-action approval policy.

What it can write

Workspace files and code through available tools · Browser clicks, typing and uploads · Host files and shell through separately enabled A0 CLI access

Surfaces

Web UI chat · Linux desktop and browser in Canvas · Optional A0 CLI Connector to the same running agent

Export / lock-in

Download generated files through the file browser and create Backup & Restore archives for chats, projects, memory, settings and workspace files. Keep a separate secure copy of secrets; memory still needs review and cleanup.

03 / Receipts

Evidence ledger

Here is what this brief is built on. “Documented” means we found a source. “Tested” means this desk actually ran it. Those are different things.

Evidence status

documented

Source set

7 primary sources

Evidence review

2026-10-15

Sources on file

Corrections

Not recorded yet. If we change a material claim, the correction belongs here.

Desk notes

What we found

Useful context, not a substitute for the source record above. This is not a desk test unless the evidence status says tested.

A useful documented fit for operators who want a self-hosted agent workbench, not a set-and-forget worker; not runtime-tested.

Surface

Docker runtime, Web UI and Linux desktop/browser; optional same-agent CLI bridge

Control

Operator-owned credentials; separate host file and code-execution grants

Automation

Optional cron/planned tasks, with user-owned failure review

Evidence

Upstream documentation; not runtime-tested

What the vendor says

README and guides describe an actual running workbench, project state and scheduling. Launcher and CLI are setup/access surfaces, not separate agent admissions.

GitHub

Canonical repository: agent0ai/agent-zero. Repository identity checked; stars, issue quality and core release chronology aren't assessed here.

Product Hunt

Not assessed

Sources fetched 2026-10-15

04 / Change log

Related coverage

The brief tells you what the product is. Payloads track the changes that may alter the decision.

No linked change coverage on file yet.