Agent profile

OpenClaw

Self-hosted personal AI assistant that acts across the messaging channels you already use (WhatsApp/Telegram/Slack/Signal) with shell, browser, file and automation tools.

Overview

OpenClawOverview
VendorOpenClaw (open source project; formerly clawdbot/moltbot)
LicenseMITCore Gateway is MIT; confirm license of all components
Latest version2026.7.1 2FIRST-PARTY: measured on a DigitalOcean s-1vcpu-1gb (1 vCPU / 1 GB, Ubuntu 22.04), harness run 20260826-195645, 2026-08-26. The version npm resolved as latest when our installer ran; recorded from the install log rather than read off a release page.
GitHub stars379000APPROX single-source snapshot; re-snapshot via GitHub API
LanguageTypeScript
InterfaceCLI, Desktop app, APICLI + messaging channels + Windows Hub companion app
PricingFree / OSSCore MIT; pay only model tokens

Agent capabilities

OpenClawAgent capabilities
TopologySingle agent
Planning
Tool callingMCPMCP registry + native tools
ExtensibilityMcp servers, Custom tools, Plugins
MemoryHas memory/cron/state; confirm model (Vocab 4)
Code executionNativeBash/process execution
Browser/computer useNativeBrowser is a first-class tool
MCPNativeMCP registry + local MCP mode
MultimodalText, AudioVoice wake/talk mode; canvas

Models

OpenClawModels
Model-agnosticYesSupports many providers incl. OpenAI; model failover + profile rotation
ProvidersAnthropic, OpenAI, OtherDoc-verified minimum: built-in anthropic provider (baseUrl override confirmed broken upstream, github.com/openclaw/openclaw#56679) + custom providers via the "openai-completions" protocol (any OpenAI-compatible endpoint, e.g. LM Studio, Vercel AI Gateway, NVIDIA NIM) — full built-in provider list not enumerated in this research pass
Local modelsVia proxyDoc example wires a local LM Studio server via a custom provider (baseUrl: http://localhost:1234/v1, api: openai-completions) — proxied through the OpenAI-compatible protocol, not a bundled local-inference runtime
BYO keyYes

Self-hosting

OpenClawSelf-hosting
Self-host difficultyModerateFIRST-PARTY: manual install = Node 22 + pnpm + PostgreSQL + gateway config wizard, which our production preset automates end to end. Re-checked against a measured install on 2026-08-26 (harness run 20260826-195645): automated, the whole thing completes inside our 60s polling resolution on a 1 vCPU / 1 GB box. The rating stays "moderate" because it describes doing it by hand — the steps did not get simpler, they got scripted.
DeployNative, Binarycurl install script / npm -g; confirm first-party deploy method
ScalingSingle processPer-user agent; verify
Config complexityModerateJSON5 config with several non-obvious gotchas found while building our install script: gateway.bind is an enum, not a raw host string; controlUi flags carry explicit security warnings; apiKey is a plain "${VAR}" string-interpolation reference (not a SecretRef object, as an earlier internal note wrongly assumed); overriding the built-in anthropic provider's baseUrl silently fails (upstream issue #56679), requiring a distinct custom provider instead
Min vCPU (measured)1 vCPUFIRST-PARTY: measured on a DigitalOcean s-1vcpu-1gb (1 vCPU / 1 GB, Ubuntu 22.04), harness run 20260826-195645, 2026-08-26. Installed, verified and answered a live completion on a single vCPU; load average peaked at 1.82 during the completion, so one core is a real floor rather than a comfortable one.
Min RAM (measured)1 GBFIRST-PARTY: measured on a DigitalOcean s-1vcpu-1gb (1 vCPU / 1 GB, Ubuntu 22.04), harness run 20260826-195645, 2026-08-26. Proven floor, not an estimate: every tier passed on a 1 GB box including a live model completion. Peak system use 664 MB of 961 MB usable, leaving 297 MB available — the tightest of the four agents measured. Below 1 GB is untested.
Idle RAM (measured)278 MBFIRST-PARTY: measured on a DigitalOcean s-1vcpu-1gb (1 vCPU / 1 GB, Ubuntu 22.04), harness run 20260826-195645, 2026-08-26. Resident set of the openclaw-gateway process itself 30s after a completion, not whole-system use (the box totalled 630 MB with the OS and our agent). Peaked at 344 MB during the completion. Note this figure tracks the box: on a 4 GB tier the same build settled at 293 MB and peaked at 427 MB, because the Node heap sizes itself against available RAM.
Infra costLowFIRST-PARTY: measured on a DigitalOcean s-1vcpu-1gb (1 vCPU / 1 GB, Ubuntu 22.04), harness run 20260826-195645, 2026-08-26. Runs on the cheapest tier we provision, but with the least headroom of the four CLI agents measured — a resident gateway process of ~278 MB plus a peak of 344 MB. Cheap to host, not free to host.
One-click on AgentOceanYesFIRST-PARTY: production install preset in src/lib/products.ts (id: openclaw) — provisions Docker + PostgreSQL + OpenClaw at checkout

Project health

OpenClawProject health
MaturityBetaNew (Jan 2026), rapidly maturing
MaintenanceActive
Release cadenceIrregularFrequent calendar releases
DocsGoodEstimate; audit
CommunityGrowingFast-growing since Jan 2026 launch

A dash (—) marks a value we have not yet verified first-hand. Record last verified 2026-06-17.

Run OpenClaw on your own VPS — one click

OpenClaw is a one-click install on any AgentOcean VPS: pick a plan, tick the preset, and it's live in minutes.

OpenClaw head-to-head