← All agent and harness briefs
Agent runtimeagenticcodingopsdocumentedunattended capable

Operational brief

OpenBot

Local-first Electron workspace for persistent provider-backed agents, queues and delegation. A full-access development preview, not an agent filesystem sandbox; current code is noncommercial source-available.

No product visual on file

Choose it when

You want human-assigned coding or ops work with persistent agents, channels and delegation, and accept the preview's full local access. Check the noncommercial license and provider terms before choosing it for business or shared use.

Skip it when

You need OSI open source, unrestricted commercial use without a separate license, per-action tool confirmations, an agent filesystem sandbox, or a guaranteed offline/no-egress setup.

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
Normally a person starts or assigns a session/task; agentic describes that configured work, not a per-action safety gate. Unattended=true applies only to configured webhook routines while the host is online, not an always-awake default. Host admins manage routines and secrets; external event data isn't authority to escalate permissions.
State
Independent provider sessions and persistent per-agent workspaces/conversations, with SQLite event history and queues. Provider switching keeps a local conversation, not a portable native provider session or a tested recovery guarantee.
Tools
Provider-backed agents can read/change files, run commands, use network and control the embedded browser. Queues, messages, delegation and file transfers make this a working agent harness, not just a message bus. Access can extend beyond workspace folders when OS/provider allow.
Human control
Explicit first-launch consent, then documented unrestricted execution without per-action confirmations. Channel Stop, Resume and Reassign controls don't establish tool approval gates. The operator must constrain tasks, credentials and OS access, keep backups and own stopping/recovery; sandboxed renderer/browser components don't sandbox agent filesystem actions.

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
Desktop-first Electron app on your own computer. Separate Linux server mode runs under Xvfb/systemd; self-hosted is the directory's own-host runtime category, not an IDE label. Optional vendor-managed servers are a separate offering, not assumed identical or always awake. Provider inference is separate from hosting the workspace; platform support here is documentary, not an installation test.
Data boundary
Prompts reach selected model providers; browser pages and plugins use network services. Optional accounts/remote access add hosted account and connection services. Local storage and sandboxed browser/renderer components don't establish no-egress, zero telemetry, retention guarantees or policy compliance; the privacy policy wasn't reviewed.
Identity & permissions
Provider CLIs use your own account or key; OpenBot accounts and host membership are separate. OpenBot's document says Anthropic hasn't confirmed subscription sign-in permission and warns about team requests using one person's plan. Check provider terms/API options. Server agents run as the service user with access to its files, keyring and database; webhook management belongs to host admins, not incoming event data.
What it can write
Host files and commands, potentially outside agent workspaces · Agent messages, delegated channel tasks and file transfers · Embedded browser actions
Surfaces
Electron desktop workspace · Separate Linux server mode under Xvfb/systemd · Optional desktop, iPhone and web remote clients
Export / lock-in
README documents local workspaces, SQLite history and provider-owned session stores. A stable local conversation can remain when switching providers; that isn't portable native provider state. Complete backup/restore and cross-provider replay weren't tested here. Keep backups, but don't read this as a verified round trip.

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
5 primary sources
Evidence review
2026-10-25

Corrections

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

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.