Any knowledge worker, not just engineers. Chat tab for researchers, PMs, ops, legal, marketing, finance — natural-language prompts, file access, knowledge-base lookup, drafting. Terminal tab for engineers who want the raw REPL. Same engine, same sessions, same config — different preferred surface.

Open standards, not a walled garden. Built on the conventions the agent-tooling industry is converging on, not bespoke formats you have to learn only for us. Model Context Protocol for tool servers. AGENTS.md for project instructions — the vendor-neutral standard adopted by Google, OpenAI, Factory, Sourcegraph, and Cursor. SKILL.md with YAML frontmatter for packaged workflows. Your configuration is portable between thClaws, other agents that speak the same standards, and whatever comes next.

Skills. Reusable expert workflows packaged as a directory with SKILL.md plus optional scripts. The agent picks the right skill automatically when a request matches the whenToUse trigger, or you can invoke one explicitly as /<skill-name>. Install from a git URL or .zip archive with /skill install.

MCP servers. Plug in tools built by third parties — GitHub, filesystems, databases, browsers, Slack, and more. Both stdio and HTTP Streamable transports, with OAuth 2.1 + PKCE for protected servers. Add one with /mcp add or ship a .mcp.json in your project.

Plugin system. Skills + commands + agent definitions + MCP servers bundled under a single manifest, installable from git or .zip. One install, one uninstall, one version to pin — ideal for sharing a team's extensions.

Memory & project instructions. Drop an AGENTS.md (or CLAUDE.md) in your repo — thClaws walks up from cwd and injects every match into the system prompt. A persistent memory store holds longer-lived facts the agent has learned about you, classified as user / feedback / project / reference and stored as markdown you can read, edit, or commit.

Knowledge bases (KMS). Per-project and per-user wikis the agent can search and read on demand. Drop markdown pages under .thclaws/kms/<name>/pages/, give each a one-line entry in index.md, and the agent gets a table of contents every turn plus KmsRead / KmsSearch tools. No embeddings — grep + read, following Andrej Karpathy's LLM-wiki pattern.

Agent orchestration. Delegate subtasks to isolated sub-agents via the Task tool — each gets its own tool registry and can recurse up to 3 levels deep. Scale further with Agent Teams: multiple thClaws processes coordinating through a shared mailbox and task queue, each in its own tmux pane and optional git worktree. One agent writes your backend while a teammate builds the frontend in parallel, lead merges the branches when both are done.

Settings as one file. Every knob — permission mode, thinking budget, allowed/disallowed tools, provider endpoints, KMS attachments — lives in .thclaws/settings.json (project) or ~/.config/thclaws/settings.json (user). API keys go in the OS keychain by default (macOS Keychain / Windows Credential Manager / Linux Secret Service) with .env fallback for CI.

Safety first. A filesystem sandbox scopes file tools to the working directory. Destructive shell commands are flagged before execution. You approve every mutating tool call unless you've opted into auto-approve.

Offline-capable. Ollama (native and Anthropic-compatible) lets you run entirely against a local model — no cloud round-trip, no API key.

Deploy what you build. Ship the landing pages, web apps, APIs, and AI agents you create through Agentic Press Hosting (partnered with SIS Cloud Service and Artech.Cloud) — or any other host you prefer. Schedule agents on cron, respond to webhooks, stream from public URLs. The deploy flow ships as a plugin (/plugin install …-deploy), so hosts are swappable; the client never locks you in.

Shell escape. Prefix any REPL line with ! to run a shell command directly — no tokens, no approval prompt, no agent round-trip (! git status, ! ls, etc.).