#Learning openclaw need help
1 messages · Page 1 of 1 (latest)
That third box is the thinking/reasoning level picker.
Default (adaptive) means:
- you have not manually overridden thinking for this session
- OpenClaw is using the model's resolved default
- for Claude 4.6 on Anthropic, that default is adaptive
So “adaptive” basically means the model/provider decides how much extra reasoning to use, instead of forcing off, low, medium, etc.
One useful detail: if you change that dropdown, it becomes a session override, not just a one-message setting.
Docs:
If you want, I can also explain what the practical difference is between off, low, high, and adaptive.
Can you explain all the features, categories, options, literally everything in openclaw web, TUI and all and also from https://docs.openclaw.ai/
I can teach you OpenClaw, but “literally everything” is too much for one Discord reply. It’s a pretty big system.
Best way is: I give you the map first, then we go section by section.
The big picture
OpenClaw is basically made of these layers:
-
Gateway
- the always-on backend
- handles sessions, models, tools, channels, auth, config, automation
-
Chat surfaces
- Web / Control UI
- TUI (
openclaw tui) - chat apps like Discord, Telegram, WhatsApp, etc.
-
Agents
- named assistants like
main,research, etc. - each has prompts, tools, routing, defaults
- named assistants like
-
Sessions
- each conversation lives in a session
- sessions can have overrides like model, thinking, verbose, trace, delivery
-
Models
- OpenAI, Anthropic, local models, etc.
- model selection + fallback + provider auth
-
Tools / skills / plugins
- file read/write, web search, browser, messaging, PDFs, images, exec, etc.
- skills add workflows and specialized behavior
-
Automation
- cron jobs
- webhooks
- recurring tasks
-
Nodes / media / voice
- camera, audio, mobile/device integrations, voice workflows
-
Security + pairing
- gateway auth
- device approval
- exec approvals
- remote access rules
-
Config + troubleshooting
openclaw.json- doctor, logs, health, status, troubleshooting docs
Web UI / browser stuff
There are 3 separate things people mix up a lot:
1. Control UI
This is the main browser control panel.
Docs: https://docs.openclaw.ai/web/control-ui
It can do things like:
- chat
- switch model / thinking
- list sessions
- patch session settings
- view/edit config
- channels status/login
- skills
- cron jobs
- nodes
- logs
- health/debug info
- updates
2. WebChat
This is the chat experience over the gateway websocket.
Docs: https://docs.openclaw.ai/web/webchat
Think of it as:
- the chat surface itself
- same sessions/routing rules
- chat history + send + abort + inject
3. Dashboard / web gateway access
This is the web entry point / browser-facing layer around the gateway.
Docs: https://docs.openclaw.ai/web
TUI
The TUI is the terminal UI:
- start with
openclaw tui - lets you chat from terminal
- has slash commands
- session/agent/model pickers
- delivery toggle
- can show tool cards/output
Docs:
Key TUI ideas:
/agent,/session,/model/think,/fast,/verbose,/trace,/reasoning
/deliver on|off!commandruns a local shell command on the TUI host
Best docs path if you're new
I’d go in this order:
-
Getting started
https://docs.openclaw.ai/start/getting-started -
Docs directory
https://docs.openclaw.ai/start/docs-directory -
Docs hubs
https://docs.openclaw.ai/start/hubs -
Control UI
-
Configuration
https://docs.openclaw.ai/gateway/configuration -
Troubleshooting
https://docs.openclaw.ai/gateway/troubleshooting
If you want the shortest mental model
- Gateway = brainstem/server
- Control UI = browser control panel
- WebChat/TUI/Discord/etc. = places you talk to it
- Agent = which assistant/persona/config you’re using
- Session = which conversation memory thread you’re in
- Model = which LLM is answering
- Thinking = how much reasoning effort it uses
- Tools = what actions it can take
- Config = how everything is wired together