Overview

Relevant source files

• .agents/skills/clawsweeper/SKILL.md
• .agents/skills/openclaw-landable-bug-sweep/SKILL.md
• .agents/skills/openclaw-pr-maintainer/SKILL.md
• .agents/skills/openclaw-small-bugfix-sweep/SKILL.md
• .agents/skills/openclaw-test-performance/SKILL.md
• .agents/skills/openclaw-testing/SKILL.md
• .github/instructions/copilot.instructions.md
• .github/labeler.yml
• .github/pull_request_template.md
• .github/workflows/openclaw-release-telegram-qa.yml
• AGENTS.md
• CONTRIBUTING.md
• README.md
• docs/.i18n/glossary.zh-CN.json
• docs/AGENTS.md
• docs/assets/openclaw-banner-dark.png
• docs/assets/openclaw-banner-light.png
• docs/concepts/model-providers.md
• docs/docs.json
• docs/docs_map.md
• docs/install/daytona.md
• docs/plan/swarms.md
• docs/platforms/chromeos.md
• docs/platforms/index.md
• docs/plugins/reference.md
• docs/plugins/reference/anthropic.md
• docs/plugins/reference/clawrouter.md
• docs/plugins/reference/deepseek.md
• docs/plugins/reference/featherless.md
• docs/plugins/reference/github-copilot.md
• docs/plugins/reference/google.md
• docs/plugins/reference/minimax.md
• docs/providers/index.md
• docs/providers/models.md
• docs/providers/moonshot.md
• docs/releases/2026.6.11.md
• docs/releases/index.md
• docs/tools/index.md
• docs/vps.md
• extensions/kimi-coding/implicit-provider.test.ts
• extensions/kimi-coding/index.test.ts
• extensions/kimi-coding/index.ts
• extensions/kimi-coding/kimi-coding.live.test.ts
• extensions/kimi-coding/provider-catalog.test.ts
• extensions/kimi-coding/provider-catalog.ts
• extensions/kimi-coding/provider-policy-api.test.ts
• extensions/kimi-coding/provider-policy-api.ts
• extensions/kimi-coding/stream.test.ts
• extensions/kimi-coding/stream.ts
• extensions/telegram/src/target-writeback.test-shared.ts
• extensions/telegram/src/target-writeback.ts
• scripts/README.md
• scripts/check-deprecated-api-usage.mjs
• scripts/docs-list.d.ts
• scripts/e2e/openwebui-probe.mjs
• scripts/generate-plugin-inventory-doc.mjs
• scripts/github/barnacle-auto-response.mjs
• scripts/github/real-behavior-proof-check.mjs
• scripts/github/real-behavior-proof-policy.mjs
• scripts/lib/deprecated-plugin-sdk-usage.d.mts
• scripts/lib/deprecated-plugin-sdk-usage.mjs
• scripts/lib/direct-run.mjs
• scripts/lib/regexp.mjs
• scripts/pre-commit/pnpm-audit-prod.mjs
• scripts/release-telegram-qa.mjs
• skills/gh-issues/SKILL.md
• src/channels/turn/durable-delivery.test.ts
• src/channels/turn/durable-delivery.ts
• src/test-utils/channel-plugins.ts
• test/e2e/qa-lab/runtime/openwebui-probe.e2e.test.ts
• test/scripts/barnacle-auto-response.test.ts
• test/scripts/check-deprecated-api-usage.test.ts
• test/scripts/openclaw-release-telegram-qa-workflow.test.ts
• test/scripts/pnpm-audit-prod.test.ts
• test/scripts/real-behavior-proof-policy.test.ts

OpenClaw is a multi-platform AI gateway and personal assistant system that orchestrates conversations between messaging channels and AI agents README.md18-19 It acts as a central control plane, connecting messaging platforms like WhatsApp, Telegram, Slack, Discord, Signal, and iMessage to advanced AI models, providing a unified interface for AI interaction README.md4-7 docs/docs.json4

The system is powered by the Pi Agent Core docs/docs_map.md105-125 and includes the Codex app-server harness for specialized coding and tool-use workloads docs/docs_map.md203-214 It is designed around a Personal Assistant Trust Model, assuming a single trusted operator boundary. It enables agents to perform real-world tasks through a rich tool ecosystem, including browser automation, filesystem access, and media understanding README.md58-62 AGENTS.md50-51

For installation and first-time setup, see Getting Started. For fundamental mental models, see Core Concepts. For technical details on the build system and cross-platform clients, see Platform Architecture.

Sources: README.md18-19 docs/docs.json4 docs/docs_map.md105-125 docs/docs_map.md203-214 AGENTS.md50-51

Core Architecture

OpenClaw is organized as a monorepo. The Gateway is the central engine that manages the lifecycle of the assistant, configuration validation, and agent execution.

System Mapping: Natural Language to Code Entities

Sources: README.md56-62 docs/docs_map.md22-29 docs/docs_map.md203-214 docs/docs_map.md29-38 docs/docs_map.md295-305

Core Concepts

Gateway & Agents

The Gateway is the always-on control plane. It manages Agents, which are execution contexts configured with specific models and workspace defaults. The system maintains a synchronized model catalog for discovery, including support for providers like OpenAI, Anthropic, and Google Gemini docs/concepts/model-providers.md10-18 Root rules for agent behavior and repository-wide standing orders are defined in AGENTS.md AGENTS.md1-5

Sessions & Channels

A Session represents a specific conversation thread. Channels are transport layers that route messages to the Gateway README.md58-59 OpenClaw supports a wide array of integrations, including Telegram, Discord, Slack, and Signal docs/docs.json4 Recent updates emphasize robust message delivery and session management across these platforms.

Skills & Tools

Tools are functional capabilities (like shell execution, web search, or browser automation) that the agent can invoke README.md61-62 Skills are modular extensions that teach agents how to use these tools for specific domains, managed via SKILL.md files such as the PR maintainer skill .agents/skills/openclaw-pr-maintainer/SKILL.md1-4

For a deeper dive into these primitives, see Core Concepts.

Sources: docs/concepts/model-providers.md10-18 README.md58-62 AGENTS.md1-5 .agents/skills/openclaw-pr-maintainer/SKILL.md1-4

Multi-Platform Architecture

OpenClaw is built for high portability across desktop, server, and mobile environments. It leverages a shared TypeScript core and native UI components.

Platform	Role	Technology
Server / Desktop	Core Gateway	Node.js (>=24.15) / TypeScript AGENTS.md50-51 scripts/pre-commit/pnpm-audit-prod.mjs8-9
Web	Control UI	React/Lit (via ui/ workspace) README.md56-57
Native Clients	iOS / Android / macOS / Linux	Swift / Kotlin / Tauri (via apps/) docs/docs_map.md315-330
Container	Deployment	Docker Multi-arch manifests docs/docs_map.md345-350

Data Flow: Message to Tool Execution

Sources: AGENTS.md1-5 README.md56-62 docs/docs_map.md340-350 docs/docs_map.md130-145

For details on the build system and native integrations, see Platform Architecture.

Security & Trust

OpenClaw operates on a Personal Assistant Trust Model. Because the assistant has access to sensitive local tools (like shell execution), it includes several safety boundaries:

• Security Boundaries: Strict controls on filesystem access and repo-root references AGENTS.md9
• Sandboxing: Docker-based isolation or remote backends like OpenShell for tool execution docs/docs_map.md340-350
• Audit System: Diagnostic repair and security hardening via the doctor command docs/docs_map.md26-27
• Repair Doctrine: Architectural root-cause investigation is required for all fixes to ensure the system remains robust AGENTS.md23-28

Sources: AGENTS.md9 AGENTS.md23-28 docs/docs_map.md340-350 docs/docs_map.md26-27

Getting Started

The recommended way to set up OpenClaw is via the interactive onboarding wizard.

This command guides you through:

1. Model Providers: Configuring API keys for providers like OpenAI, Anthropic, or Google Gemini docs/concepts/model-providers.md15-20
2. Messaging Channels: Linking accounts for Telegram, WhatsApp, Discord, and others README.md58-59
3. Workspace Setup: Initializing the configuration and agent workspaces README.md52-53

For a full walkthrough, see Getting Started.

Sources: README.md46-53 docs/concepts/model-providers.md15-20