Self-hosted

BoltaClaw CLI

Run Bolta's AI social media agents on your own machine while controlling them from the Bolta dashboard. Your data, API keys, and drafts stay local — Bolta Cloud handles scheduling, OAuth, and publishing.

BoltaClaw ships with bolta-skills built in

36+ skills across 8 planes — your agents know how to draft, schedule, review, and respond out of the box. Reinstall or update via boltaclaw oc skills install bolta-skills-index, or run boltaclaw update to refresh everything at once.

Architecture

BoltaClaw is a thin bridge layer over OpenClaw, an open-source agent runtime. Your machine runs the work; Bolta Cloud is the control tower.

Bolta Cloud

Control tower

• Dashboard / Conductor UI
• Job queue and scheduler
• Social OAuth tokens
• WebSocket relay

Your Machine

The factory (outbound WSS)

• OpenClaw daemon
• bolta-skills
• Bolta MCP client
• Local SQLite (memory + drafts)
• Your Claude API key (BYOK)
• Optional Telegram / Slack / Discord bots

Quick start

Get a workspace install token from Settings → Integrations (Self-Hosted / engine). Then install using one of the options below.

bash

curl -sL https://bolta.ai/install.sh | bash -s -- --token=YOUR_WORKSPACE_TOKEN

Authentication

Option 1: Install token (pairing) — paste the one-time or workspace token from the dashboard when you run boltaclaw setup or boltaclaw start --token=…. After a successful handshake, Bolta issues a persistent runner credential stored locally (check boltaclaw paths).

Option 2: Bolta API key — for calling Bolta APIs from skills, set BOLTA_API_KEY (see the bolta-skills README). Keys never ship in docs — generate under Settings → API.

Where things live

OpenClaw resolves skills and agent state under ~/.openclaw (bolta profile), workspace skills under <workspace>/skills, and optional extra dirs from config — same precedence as documented in the Skills section below.

How it works

1
Install the engine on your machine (Mac, Linux, Windows WSL, Docker, VPS — anywhere).
2
Install script sets up OpenClaw — the open-source agent runtime that handles LLM calls, tool execution, memory, scheduling, and channel integrations.
3
BoltaClaw bridges OpenClaw to Bolta Cloud over an outbound WebSocket — no port-forwarding needed.
4
Bolta dashboard detects the connection and lights up green.
5
Configure agents from the Bolta UI — voice, schedule, social accounts.
6
Jobs run locally through OpenClaw — LLM calls use your own API key, drafts stay on your machine until you approve.

What runs where

LLM calls (Claude / GPT)

Your machine

BYOK — your key, your cost control.

Agent memory

Your machine (SQLite)

Privacy — brand voice, drafts, history stay local.

Social OAuth tokens

Bolta Cloud

Security — managed OAuth flow.

Job scheduling

Bolta Cloud → pushed to runner

Convenience — set schedules from the UI.

Post approval

Bolta Cloud (Inbox)

Workflow — review before publishing.

Publishing

Bolta Cloud (via social APIs)

Reliability — managed API connections.

Skills

Skills are AgentSkills-compatible directories that teach an agent how to use a tool — each is a folder with a SKILL.md and YAML frontmatter. BoltaClaw ships the bolta-skills pack bundled, so your agents know Bolta workflows out of the box.

Install bolta-skills

The pack is published to ClawHub. The install script and boltaclaw onboard grab it automatically — use the command below to reinstall or pull updates.

bash

# Install (or reinstall) the bolta-skills pack
boltaclaw oc skills install bolta-skills-index

# Update OpenClaw + bolta-skills together
boltaclaw update

Browse the registry entry at clawhub.ai/maxfritzhand/bolta-skills-index or read the source at github.com/boltaai/bolta-skills.

The 8 skill planes

bolta-skills is organised into planes — logical groupings that separate concerns so agents can compose capabilities cleanly.

Voice

4 skills

Brand voice creation, evolution, and validation.

bolta.voice.bootstrapbolta.voice.learn_from_samples

Content

8 skills

Content creation, planning, and workspace context.

bolta.draft.postbolta.week.planbolta.get_voice_profile

Review

6 skills

Human-in-the-loop approval and feedback workflows.

bolta.inbox.triagebolta.approve_postbolta.review.digest

Automation

2 skills

Scheduled and recurring content generation.

bolta.cron.generate_to_reviewbolta.cron.generate_and_schedule

Agent

6 skills

Agent lifecycle — hire, configure, activate, memory.

bolta.agent.hirebolta.agent.memorybolta.job.execute

Analytics

3 skills

Performance metrics, audience insights, posting times.

bolta.get_post_metricsbolta.get_best_posting_times

Engagement

3 skills

Mentions, comments, on-brand replies.

bolta.get_mentionsbolta.reply_to_mention

Control

6 skills

Governance, audit, key rotation, quotas.

bolta.audit.export_activitybolta.team.rotate_keybolta.workspace.config

Full skill index Skills reference (OpenClaw)

Where skills live

OpenClaw resolves skills in this precedence order — same name in multiple places, highest source wins.

<workspace>/skillsWorkspace

<workspace>/.agents/skillsProject agent

~/.agents/skillsPersonal agent

~/.openclaw/skillsManaged / local

shipped with the installBundled (incl. bolta-skills)

skills.load.extraDirs (config)Extra dirs

Commands

Prefer boltaclaw commands for normal operations. Use boltaclaw openclaw … (or boltaclaw oc …) for advanced OpenClaw utilities not yet wrapped — passthroughs always run with --profile bolta.

Lifecycle

Start the engine, check its health, restart when something drifts.

bash

boltaclaw start --token=YOUR_TOKEN

Start the engine and connect to Bolta Cloud.

bash

boltaclaw status [--json]

Show OpenClaw, gateway, connection, and key status.

bash

boltaclaw whoami

Show workspace status and connected agents.

bash

boltaclaw restart

Restart the local gateway in one step.

bash

boltaclaw gateway health

Check the local gateway health (alias: boltaclaw gw health).

bash

boltaclaw gateway start

Start the local gateway only.

bash

boltaclaw gateway stop

Stop the local gateway.

Setup & onboarding

First-run wizards that pair your machine with a workspace.

bash

boltaclaw setup [--token=YOUR_TOKEN]

Interactive setup wizard.

bash

boltaclaw onboard --token=YOUR_TOKEN

Guided onboarding (includes optional OpenClaw native onboard).

Run agents

Send a prompt or hold a conversation with your agents from the terminal.

bash

boltaclaw run

Run an agent with a prompt (API-based, server-side execution).

bash

boltaclaw chat

Interactive chat with your agents (API-based).

bash

boltaclaw action --help

Inspect agent actions exposed via OpenClaw.

Manage agents & jobs

Hire, enable, and inspect agents and the jobs that drive them.

bash

boltaclaw agent list

List hired agents in this workspace.

bash

boltaclaw agent hire

Hire an agent preset (creates agent + default job).

bash

boltaclaw agent enable <id>

Activate an agent.

bash

boltaclaw agent disable <id>

Pause an agent.

bash

boltaclaw job list

List jobs across hired agents.

bash

boltaclaw job create

Create a job for an agent.

bash

boltaclaw job enable <id>

Activate a paused job.

bash

boltaclaw job disable <id>

Pause a running job.

bash

boltaclaw presets

Browse available agent presets.

Configuration

Locally stored settings — BYOK keys, model selection, optional channel tokens.

bash

boltaclaw config set ANTHROPIC_API_KEY sk-ant-...

Set your Claude API key. BYOK — never leaves your machine.

bash

boltaclaw config set MODEL_PRIMARY anthropic/claude-sonnet-5

Pick the primary model used by agents.

bash

boltaclaw config set TELEGRAM_BOT_TOKEN 123456:ABC...

Optional — wire up a Telegram bot channel.

bash

boltaclaw config get <KEY>

Print a single configured value.

bash

boltaclaw config list

List all configured values.

bash

boltaclaw config unset <KEY>

Remove a configured value.

Diagnostics & ops

Logs, paths, doctor, and an interactive ops dashboard.

bash

boltaclaw doctor [--fix]

Diagnose local setup and command failures. --fix attempts repairs.

bash

boltaclaw logs -f

Tail OpenClaw gateway logs.

bash

boltaclaw paths

Show local runtime paths (db, config, workspace).

bash

boltaclaw tui

Live ops dashboard (interactive TUI).

OpenClaw passthrough

Run any native OpenClaw command inside the isolated bolta profile (alias: oc).

bash

boltaclaw oc skills install bolta-skills-index

Install the bolta-skills pack from ClawHub.

bash

boltaclaw oc gateway health

Check OpenClaw gateway health under the bolta profile.

bash

boltaclaw openclaw uninstall

Uninstall OpenClaw from the bolta profile.

Maintenance

Keep OpenClaw and bolta-skills current; cleanly remove the engine.

bash

boltaclaw update

Update OpenClaw and bolta-skills.

bash

boltaclaw uninstall --yes

Remove the local OpenClaw bolta profile.

bash

boltaclaw uninstall --yes --purge

Also delete local BoltaClaw data.

Troubleshooting

Quick reference

bash

# 1) Fast diagnostics
boltaclaw doctor
boltaclaw doctor --fix

# 2) Gateway not healthy
boltaclaw gateway start
boltaclaw gateway health
boltaclaw logs -f
boltaclaw restart

# 3) Verify OpenClaw under bolta profile
boltaclaw oc gateway health
boltaclaw oc --help

# 4) Re-onboard if token/key drifted
boltaclaw onboard --token=YOUR_TOKEN

# 5) Full cleanup (last resort)
boltaclaw uninstall --yes --purge

Common causes of failures

Missing API key (ANTHROPIC_API_KEY or OPENAI_API_KEY).
Gateway not running on the configured local port (127.0.0.1:18789 by default).
Token / config drift after reinstall or machine migration.
Running native openclaw outside the bolta profile by accident.

Security

Outbound-only connections — your machine initiates all connections, no inbound ports.
BYOK — API keys never leave your machine.
Token burn — install tokens are single-use; a persistent runner key is issued after handshake.
End-to-end encryption — WSS (TLS) for all cloud communication.
Local storage — memory, drafts, and history live in local SQLite.

Requirements

Node.js 18+ (auto-installed by the install script).
Internet connection (for Bolta Cloud and LLM APIs).
Claude (Anthropic) or OpenAI API key.

BoltaClaw on GitHub bolta-skills API reference