Command line (agentbees)

Every AgentBees workspace ships with the agentbees CLI on PATH, ready to use with zero install and zero login. When a task starts, the platform injects a short-lived, workspace-scoped token so the CLI is already authenticated as you, scoped to your org.

The CLI is workspace-only. It runs inside a task/workspace, where the token is injected — there’s no laptop login, personal access token, or config file. Run it outside a workspace and every command fails closed with a clear message.

Nothing to set up

There’s no login, no API key, and no endpoint to configure. Because you’re already signed in and the task runs in your workspace, agentbees is authenticated automatically as you, scoped to your org — just run a command. The server enforces all permissions (row-level security + your role); the CLI only makes API calls on your behalf.

Under the hood (you don't need this)

When a task starts, the platform injects two environment variables into the pod — the CLI reads them so you never touch them:

AGENTBEES_TOKEN — a short-lived, workspace-scoped bearer token, minted per task.
AGENTBEES_API_URL — this deployment’s own API origin, so the same binary always targets the right app.

These are set for you; you never export or manage them.

Command reference

Every command supports --json for scripting.

Command	What it does
`agentbees whoami`	Your identity, org, role, and plan
`agentbees usage [--days N]`	Compute + token usage (add `--watch` to live-refresh)
`agentbees skills list`	Your org’s HoneyBox skills
`agentbees skills install <name>`	Pull a skill into `.agentbees/skills/<name>.md`
`agentbees skills push <file>`	Create/update a skill from a local file
`agentbees skills capture <task>`	Capture a task’s work as a draft skill
`agentbees models`	The active LLM + which agent CLIs you can connect to
`agentbees connect <agent>`	Launch an agent CLI (`claude` \| `codex` \| `gemini` \| `kiro`)
`agentbees checkpoint list`	List this workspace’s checkpoints
`agentbees checkpoint create`	Snapshot this workspace (`--label` optional)
`agentbees restore <id>`	Restore a checkpoint (in place, or `--recover` into a fresh pod)
`agentbees hive`	The agents in this project’s hive
`agentbees hive conflicts`	Cross-branch conflicts vs your siblings
`agentbees hive diff <task>`	Pairwise diff vs a sibling (overlapping paths)

Identity & usage

agentbees whoami          # who am I, which org, which plan
agentbees usage           # this period's usage + cost (last 30 days)
agentbees usage --days 7  # a shorter window
agentbees usage --watch   # live-refresh every 5s

Skills (HoneyBox)

Pull reusable org playbooks into the workspace, or contribute new ones. See HoneyBox skills for the library itself.

agentbees skills list
agentbees skills install migration-runbook   # → .agentbees/skills/migration-runbook.md
agentbees skills push ./playbook.md --name deploy-checklist --tags deploy,ops
agentbees skills capture <task-id>            # turn a good run into a draft skill

Models & agents

models shows your org’s active LLM plus, for each bundled agent CLI, whether it’s available and provisioned in this task. connect launches one interactively.

agentbees models
agentbees connect claude   # drops you into the agent CLI, using the task's creds

connect uses the credentials the platform already injected for this task. Connecting an agent that isn’t provisioned here still launches — it may prompt for its own sign-in.

Checkpoints & restore

Snapshot the workspace and restore it later. Creating a checkpoint requires the backup subscription (enable it in Admin → Billing); without it the command returns a clear message.

agentbees checkpoint create --label "before the refactor"
agentbees checkpoint list
agentbees restore <checkpoint-id>            # in place — confirms first
agentbees restore <checkpoint-id> --recover  # into a fresh pod
agentbees restore <checkpoint-id> --yes      # skip the confirmation prompt

restore (in place) overwrites the current workspace changes. It prompts for confirmation on a TTY; pass --yes in non-interactive shells.

The hive

Read-only situational awareness for coordinated work — who else is in the hive, and how your branch relates to theirs.

agentbees hive                 # roster: agent, role, status, branch (marks "you")
agentbees hive conflicts       # ✓ clean / ✗ conflicting paths vs each sibling
agentbees hive diff <task-id>  # files touched on both branches + the overlap

The messaging tools the agents use (hive_broadcast / hive_ask / hive_inbox, the BeesBoard, cell claims) are agent-to-agent and not exposed on the CLI — a person isn’t a bee in the swarm. The CLI surfaces only the read-only views a developer wants.

Scripting

# Machine-readable output for any command
agentbees usage --json | jq '.summary.totals.costUsd'
agentbees hive --json  | jq '.agents[] | select(.isSelf|not) | .branch'

Exit codes: 0 success, 2 auth / unreachable, 1 everything else.

Getting started

Workflows

Agents & models

Command line

Billing

Enterprise

Command line (agentbees)

Nothing to set up

Command reference

Identity & usage

Skills (HoneyBox)

Models & agents

Checkpoints & restore

The hive

Scripting

​Nothing to set up

​Command reference

​Identity & usage

​Skills (HoneyBox)

​Models & agents

​Checkpoints & restore

​The hive

​Scripting

Nothing to set up

Command reference

Identity & usage

Skills (HoneyBox)

Models & agents

Checkpoints & restore

The hive

Scripting