Skip to content
Briar
Open the workbench
v2.4.0 · live preview available

The workbench for engineering AI agents that ship.

Briar lets you compose specialist agents, equip them with tools and skills, plug them into GitHub, Jira, and Datadog, and watch them triage tickets, draft pull requests, and ship to production — all from a live kanban board.

Open the workbenchSee how it works
Agents
12+
Native integrations
6
Tasks shipped / wk
2.4k
briar · workbenchOnline
$ briar agents deploy backend-bug-fixer
› binding tools: shell_exec, db_inspector, github_pull
› subscribed to triggers: github.issues.opened, datadog.alert
✓ agent online · awaiting tasks
TSK-9417 routed to backend-bug-fixer · 14:02:11
tool_call analyze_heap → leak_probability=0.94
patch drafted · 142 additions / 28 deletions
✓ PR #842 opened · awaiting review
$

Plugs into the tools your engineers already use

  • GitHub
  • Jira
  • Datadog
  • Linear
  • AWS S3
  • Slack
  • Sentry
  • PagerDuty

What Briar gives you

One workbench for the full agent lifecycle.

  • Compose

    Design specialist agents from a clean form: identity, system prompt, model, tools. Wrap them in skills — composable step chains that turn an intent into a deterministic action.

  • Run

    Triggers turn webhooks, cron ticks, and monitor alerts into tasks. Workflows route them through the right agents. The kanban board shows every step, live, with full reasoning traces.

  • Observe

    Cost, latency, and tokens per agent in one dashboard. A smart debugger surfaces what to fix next. Every secret read and tool call is recorded in the audit log.

The primitives

Eight primitives. Endless agent topologies.

Briar is opinionated about the building blocks and quiet about everything else. Each primitive is independently editable, version controlled, and observable.

  • Agents

    Specialist AI workers with their own persona, model, temperature, and toolset. Bug-fixers, log analyzers, dependency upgraders — composed from the same primitives.

    Browse agents

  • Tools

    Reusable verbs your agents can call. Strict input/output schemas, side-effect policy, and a versioned registry — so every call is reviewable and reproducible.

    Open tool builder

  • Skills

    Composable step chains. Stitch a prompt template, a search step, and a patch step into a deterministic action that any agent can invoke.

    Design a skill

  • Workflows

    Visual node graph editor. Wire triggers to routers to agent actions, with branches by severity, environment, or any payload field.

    Open the editor

  • Triggers

    Map external event payloads to task fields with JSONPath. GitHub issues, Datadog alerts, cron ticks, custom webhooks — all become work the kanban board picks up.

    Configure triggers

  • Sources

    Connect GitHub, Jira, Datadog, Linear, AWS S3 and more. Scoped credentials, last-sync telemetry, and one-click reconnects when auth drifts.

    View sources

  • Runs & observability

    Every execution is a first-class object: run-log timeline, tool calls, token usage, cost. Smart-debugger insights tell you what to optimize next.

    See the dashboard

  • Secrets vault

    Encrypted at rest, rotated automatically, scoped per environment. Every read is recorded against the agent and run that requested it.

    Open settings

How it works

Three steps from a fresh repo to autonomous engineering.

  1. 01

    Connect your sources

    Authorize GitHub, Jira, Datadog, Linear, or any webhook. Scopes are minimal and rotatable.

    ›github · connected
    ›jira · connected
    ›datadog · syncing
  2. 02

    Compose a specialist agent

    Pick a model, write a system prompt, attach the tools and skills it’s allowed to call. Test it in the live playground.

    ›model: claude-3-5-sonnet
    ›tools: shell_exec, github_pull
    ›skill: patch-and-test
  3. 03

    Bind a trigger and let it ship

    Map an event payload to task fields. Tasks land on the kanban board and run through the workflow you designed.

    ›trigger: github.issues.opened
    ›→ task.title = $.issue.title
    ›→ workflow: incident-triage

Live workbench

Watch every reasoning step, in real time.

The kanban board on the left is the source of truth for what your agents are doing. Click any card and the task detail panel exposes the full run log — every tool call, every output, with human checkpoints when an action touches critical infrastructure.

In progress 2

84.2k tok
Refactor authentication middleware
BB
WRITING TESTS…
68.1k / $0.22
04:12
Backfill telemetry on legacy job runners
DL
INSTRUMENTING…
16.1k / $0.06
02:04
PRJ-842Fix Memory Leak in Auth Module
In Progress
Parsed issue description & attached logs00:02s
tool_call · analyze_heap04:15s
{
  "status": "anomaly_detected",
  "leak_probability": 0.94,
  "suspect_path": "src/auth/SessionManager.ts:142",
  "retained_bytes": "45.2MB"
}
Human Checkpoint Reachedawaiting

Agent intends to modify src/auth/SessionManager.ts. This file is flagged as critical infrastructure.

Security & control

Production-grade guardrails, designed in.

  • Encrypted by default

    AES-256-GCM at rest. Per-environment scoping. Automatic 90-day rotation, revocable in one click.

  • Auditable, not opaque

    Every secret read, tool call, and file touched is recorded against the agent and run that requested it.

  • Human checkpoints

    Tag any path or table as critical infrastructure. Agents pause, surface the diff, and wait for approval before writing.

  • Bring your own keys

    Use your own Anthropic, OpenAI, and cloud credentials. No prompts or completions are stored on Briar infrastructure.

Spin up an agent in under five minutes.

The workbench ships with a working incident-triage workflow. Open it, point it at a repo, and let an agent draft its first PR.

Open the workbenchRead the docs