Is OpenClaw Free? Real Pricing, Plans & Limits (2026)
Open-source AI agent with 382,545 GitHub stars, runs locally, controls your browser and files via WhatsApp, Slack, and 25+ chat apps. Free, MIT licensed.
OpenClaw is a free, MIT-licensed AI agent for developers and technical solopreneurs who want a local assistant that actually takes actions, not just chats. It runs shell commands, browses the web, and manages files and email from WhatsApp, Telegram, or Slack. Version v2026.7.1-beta.3 (July 9, 2026) added GPT-5.6 support and external harness pairing with Codex.
OpenClaw is a free, open-source autonomous AI agent that has passed 382,545 GitHub stars as of July 2026, running locally on your machine. Created by Peter Steinberger in 2025, it is now stewarded by the nonprofit OpenClaw Foundation with OpenAI sponsorship. It integrates 50+ LLMs including Claude Opus 4.5, GPT-5.6, and xAI Grok across 25+ messaging platforms, controlling your browser, files, and email from WhatsApp or Slack.
Maker: OpenClaw Foundation (OpenAI-sponsored) · Autonomy: semi autonomous · Maturity: BETA
Underlying models: Claude Sonnet 4.6, Claude Opus 4.5, GPT-4, GPT-5.5, GPT-5.6, Gemini, DeepSeek, xAI Grok, Custom (proprietary)
About OpenClaw
OpenClaw is a free, open-source autonomous AI agent created by Peter Steinberger (founder of PSPDFKit) and launched in November 2025 under the name Clawdbot. It was renamed Moltbot over an Anthropic trademark concern, then settled as OpenClaw on January 30, 2026. As of July 11, 2026 the project has passed 382,545 GitHub stars and roughly 80,000 forks, making it one of the fastest-growing open-source projects in history and, by several trackers, the most-starred non-aggregator repository on GitHub. In February 2026 Steinberger announced he was joining OpenAI. OpenClaw itself was not acquired: it now runs under the independent, nonprofit OpenClaw Foundation (a 501(c)(3), established with a full-time team in June 2026), with OpenAI providing sponsorship rather than ownership. OpenClaw runs entirely on your local machine (Mac, Windows via WSL2, or Linux) and connects to any external language model you choose. Unlike cloud-only assistants, it executes tasks autonomously: running shell commands, controlling your web browser, reading and writing files, managing your calendar, and sending emails. The agent operates from messaging platforms including WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams, and 19+ others, plus native iOS and Android apps launched in June 2026. A heartbeat daemon monitors systems and runs background tasks without prompting. The framework is model-agnostic, supporting Claude Sonnet 4.6 (86.9% on PinchBench), Claude Opus 4.5 (89% on SWE-Bench Verified when running OpenClaw workflows), GPT-4, GPT-5.6, Gemini, DeepSeek, xAI Grok, and local models via Ollama. OpenClaw suits developers automating code review and deployment, small business owners handling email and scheduling, and data analysts building autonomous pipelines. It ships 100+ pre-configured AgentSkills, persistent memory across sessions, voice control with wake words, and real-time browser automation via a Live Canvas workspace. The July 2026 release cycle added GPT-5.6 model support, an openclaw attach external harness mode for pairing with tools like Codex over Telegram, and a Skill Workshop for building custom AgentSkills. The technical barrier stays substantial: Node >= 22, WSL2 on Windows, manual model configuration, and comfort with system permissions. OpenClaw itself costs nothing under the MIT license; see the pricing FAQ below for exact OpenClaw Cloud and self-hosted numbers. Security is the open, unresolved question: a January 2026 audit found 512 vulnerabilities including 8 critical ones tied to prompt injection, memory poisoning, and credential leakage, and the disclosed-CVE count has since grown past 138, including two CVSS 9.9 flaws that allowed unauthenticated admin takeover. Third-party scans have found 135,000+ publicly exposed instances, most with no auth layer, which is why enterprise adoption remains limited despite the feature set. The latest build is v2026.7.1-beta.3, shipped July 9, 2026, adding GPT-5.6 catalog support, Telegram-based Codex pairing and steering with mid-run recovery, Ollama node auto-discovery, and a Control UI reasoning-picker refresh. For developers and power users comfortable with local hosting, OpenClaw offers autonomy and extensibility other agents do not match at any price. For enterprises and non-technical teams, the unresolved security track record and lack of SLAs still make managed alternatives the safer call.
Pricing
Free and open-source (MIT license) with all features included. OpenClaw Cloud, the managed service, costs $49/month or $39/month billed annually. Self-hosted costs vary with usage: roughly $0-10/month in VPS hosting plus $6-30 per 1M tokens of model API usage, typically $20-50/month for serious personal or small-business use, and $200+/month for heavy or enterprise workloads.
Key Features
- Local-first gateway daemon: Runs as a persistent background daemon on macOS, Linux, or Windows via WSL2, keeping your data and state off vendor clouds.
- 25+ messaging platform integrations: Works from WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams, and 19+ other chat apps, no separate dashboard required.
- Model-agnostic model routing: Supports 50+ LLMs including Claude Opus 4.5 (89% SWE-Bench Verified), GPT-5.6, xAI Grok, DeepSeek, and local Ollama models, switchable without code changes.
- Live Canvas browser automation: An agent-driven visual workspace with A2UI lets OpenClaw fill forms, click through sites, and show you what it is doing in real time.
- 100+ AgentSkills with persistent memory: Ships over 100 pre-built AgentSkills plus a heartbeat daemon for background tasks and OpenClaw-RL self-improvement across sessions.
- Voice Wake and Talk Mode: Native iOS, Android, and macOS menu bar apps support wake-word voice control and a Twilio/Gemini voice bridge for Google Meet.
- External harness attach mode: The openclaw attach command (added July 2026) pairs OpenClaw with external tools like Codex over Telegram for interactive, resumable workflows.
Strengths
- Free and open-source under the MIT license. 382,545 GitHub stars and roughly 80,000 forks as of July 11, 2026, now stewarded by the nonprofit OpenClaw Foundation with OpenAI sponsorship.
- Truly model-agnostic across 50+ LLM integrations including GPT-5.6 and xAI Grok. Claude Opus 4.5 reaches 89% on SWE-Bench Verified through OpenClaw workflows; Claude Sonnet 4.6 hits 86.9% on PinchBench.
- 25+ messaging platform integrations, plus native iOS and Android apps (launched June 2026), let you trigger tasks from WhatsApp, Telegram, Slack, or Discord without switching context.
- Local execution avoids vendor lock-in. Self-hosted setups typically run $20-50/month in API usage, versus $49/month or more for comparable managed cloud agents.
- 100+ pre-configured AgentSkills, persistent memory, Live Canvas, and OpenClaw-RL reinforcement learning support custom automation and self-improving workflows.
Weaknesses
- High technical barrier: requires Node >= 22, WSL2 on Windows, manual LLM configuration, and systems knowledge. Setup has taken 45+ minutes on macOS even for experienced developers.
- Security has not stabilized. Following a January 2026 audit that found 512 vulnerabilities (8 critical), disclosed CVEs have grown past 138 by mid-2026, including two CVSS 9.9 flaws enabling unauthenticated admin takeover, with 135,000+ exposed instances found lacking any auth layer.
- Unpredictable task execution, including unnecessary reasoning loops and mid-execution objective changes, still requires manual human review for reliability.
- Higher token consumption than direct API calls solving the same task, due to continuous reasoning loops and tool orchestration overhead.
- Not enterprise-ready: no SLAs, no comprehensive audit logging, and limited hardening for multi-tenant deployment despite the broad feature set.
Frequently Asked Questions
What is OpenClaw and what does it do?
OpenClaw is a free, open-source autonomous AI agent created by Peter Steinberger in November 2025, originally launched as Clawdbot. It runs locally on your machine and executes tasks autonomously by controlling your browser, running shell commands, reading and writing files, managing calendars, and sending emails. Unlike cloud-only assistants, OpenClaw operates from 25+ messaging platforms including Slack, WhatsApp, Telegram, and Discord. It supports 50+ LLM integrations including Claude, GPT-5.6, and local Ollama models. A heartbeat daemon keeps it running in the background so it can act without a human prompting every step. As of July 11, 2026, the project has passed 382,545 GitHub stars and is stewarded by the independent, nonprofit OpenClaw Foundation with OpenAI sponsorship.
How much does OpenClaw cost in 2026?
OpenClaw itself is free and open-source under the MIT license, with no charge to download or run. The managed cloud service, OpenClaw Cloud, costs $49/month, or $39/month if billed annually. Self-hosted deployment costs depend on infrastructure and AI model usage: expect roughly $0-10/month in VPS hosting plus $6-30 per 1M tokens of API usage. Most serious personal or small-business setups land around $20-50/month total, while heavy enterprise use can exceed $200/month. Running a local model through Ollama can cut the token cost close to zero if your hardware supports it. There are no hidden platform fees beyond server hosting and the API consumption from whichever LLM provider you choose.
What are the main features of OpenClaw?
OpenClaw's core feature is its local-first gateway daemon, which runs persistently on macOS, Linux, or Windows via WSL2 and keeps your data off vendor clouds. It connects to 25+ messaging platforms, so you can trigger and monitor tasks from WhatsApp, Telegram, Slack, or Discord. Live Canvas gives the agent a visual browser workspace for real-time automation like form-filling and site navigation. It ships 100+ pre-configured AgentSkills, persistent memory across sessions, and a heartbeat daemon that runs background tasks without prompting. Native iOS, Android, and macOS menu bar apps add voice wake-word control and remote approval of actions. The July 2026 release added GPT-5.6 support and an openclaw attach mode for pairing with external tools like Codex over Telegram.
Is OpenClaw free to use?
Yes, OpenClaw's core software is completely free under the MIT license, including all messaging integrations, AgentSkills, and Live Canvas features. There is no gated free tier or usage cap on the self-hosted software itself. The only cost is what you choose to run it on: a VPS (roughly $0-10/month) and API calls to whichever LLM you connect, typically $6-30 per 1M tokens depending on the model. Local models via Ollama can bring even the API cost to zero if you have suitable hardware. You still need to bring your own API key or local model, since OpenClaw does not include any LLM access itself. Paid tiers only apply if you opt into the separately priced OpenClaw Cloud managed service.
What are the best alternatives to OpenClaw?
Claude Code is a strong pick if you want an agent scoped to a single repository with diff review inside your IDE rather than open-ended local automation. Cursor suits developers who prefer a visual IDE experience over a messaging-first interface. Devin, from Cognition, is worth choosing if you want a managed cloud coding agent without OpenClaw's local setup requirements. SuperAGI is a better fit for enterprise teams that need multi-agent CRM-style workflows with more built-in governance. For enterprise security needs generally, low-code workflow builders or orchestration platforms like Temporal trade autonomy for auditability and SLAs. OpenClaw remains the strongest option for unrestricted local automation controlled from messaging apps.
Who is OpenClaw best for?
OpenClaw fits software engineers automating code review, testing, and deployment, and small business owners who want email and scheduling handled from WhatsApp or Slack. Data analysts building autonomous reporting pipelines and open-source hobbyists comfortable with local hosting also get strong value. Remote workers and solopreneurs use it to offload administrative tasks like invoicing and calendar management. A typical use case is a solo developer who wires OpenClaw to watch a GitHub repo, triage issues, and message a Slack channel with proposed fixes. It is not a fit for non-technical users, since setup requires Node >= 22 and systems knowledge. It also is not a fit for enterprises needing audit logging and SLAs, given the January 2026 audit's 512 identified vulnerabilities.
How do you get started with OpenClaw?
Start at github.com/openclaw/openclaw and confirm you have Node >= 22 installed; Windows users need WSL2 first. Clone the repo or use the install script, then run the setup wizard to point OpenClaw at an LLM provider (Claude, GPT-5.6, Gemini, or a local Ollama model) with your own API key. Connect at least one messaging channel, such as Telegram or Slack, using the bot token flow in the setup docs. Once the gateway daemon is running, send it a test message like checking your inbox or reading a file to confirm the connection. Expect 30-45 minutes for a first working setup even with prior developer experience, longer if you are configuring WSL2 from scratch. The official docs and Discord-adjacent community channels cover troubleshooting for common permission and auth errors.
How does OpenClaw compare to Devin in 2026?
OpenClaw is free and self-hosted under the MIT license, while Devin is a paid, fully managed cloud coding agent from Cognition with no local install required. OpenClaw supports 50+ LLMs you choose and pay for separately, whereas Devin bundles model access into its subscription. On raw coding benchmarks, OpenClaw's best results come from routing to Claude Opus 4.5, which hits 89% on SWE-Bench Verified through OpenClaw workflows, in the same range Devin has reported for comparable tasks. Devin runs in a managed sandbox with no exposure to the host machine, while OpenClaw's default install has no sandboxing and depends on the user to lock it down. Devin is the safer pick for teams that want a supported, audited product with no local security exposure. OpenClaw wins on cost and flexibility for developers willing to manage their own infrastructure and accept its documented security gaps.