Claw field notebook
last updated 2026-05-07 edit on GitHub colophon
§ 3 Connections / § 3.1 · 3 min read

Index — patterns vs packages

How to think about connections — the difference between integration patterns (this section) and marketplace packages (§4 Plugins). Both matter; they answer different questions.

What §3 covers#

§3 Connections is about the integration patterns that plug into OpenClaw — the categories of things you wire up:

  • Channels — how messages flow in and out (Slack, WhatsApp, iMessage)
  • Tools — what the agent can do (read, exec, browser, MCP, etc.)
  • Models — the brain (Claude, GPT, local Ollama)
  • Memory — what survives a restart (built-in, Honcho, QMD)

§4 Plugins is about specific marketplace packages that implement these patterns — the actual Slack Connector you install, the actual GitHub MCP server, the actual web-search skill.

Mental model: §3 is “what kinds of things plug in?” §4 is “which specific things are worth using?”

Why we split it this way#

OpenClaw’s official docs cover each category separately (channels/index, tools docs, models, memory overview). For a learner, that’s four separate sections to discover.

Our editorial choice: §3 Connections groups them so you can scan the integration surface in one place. §4 Plugins has the specific reviews when you want depth on a single package. We flag this as our umbrella term in openclaw-primitives.json so readers know it’s our framing, not OpenClaw’s.

What’s in each sub-section#

§3.2 Channels#

The 24+ messaging surfaces OpenClaw connects to. Default DM policy, allowlists, pairing flow, and how to pick your first channel. If you’re getting started, Telegram or Discord first.

§3.3 Tools#

The agent’s hands. Built-in tools (read, exec, edit, write, browser, canvas, mcp) and how they relate to skills + MCP servers. Tool policy and sandbox.

§3.4 Models#

The brain choice. Provider list, model ref format, recommended models for 2026-05, failover configuration, cost notes.

§3.5 Memory#

The agent’s notebook. Three engines (built-in / Honcho / QMD), active vs long-term memory, memory search.

A useful order to read these#

If you’ve not read any §3 page yet:

  1. Channels first — they’re the most concrete and the question you ask first (“can OpenClaw connect to my Slack?”)
  2. Models second — the cost / quality decision affects everything else
  3. Tools third — once you have a model talking to a channel, what can it do?
  4. Memory last — most people use the default (built-in) for ages before needing more

How §3 patterns connect to §4 packages#

Cross-reference table — every §3 category has §4 plugins/packages that implement it:

§3 Category§4 Examples (when published)
ChannelsSlack Connector · WhatsApp Connector · Discord bridge · iMessage native · BlueBubbles
ToolsBrowser tool variants · Custom code-exec wrappers
Models(no plugin needed — providers are first-class)
MemoryHoncho service config · QMD setup
MCP (special)GitHub MCP · Stripe MCP · Browserbase MCP · Linear MCP · Figma MCP — all reviewed in §4

When §4 Plugins ships individual field notes, each one will declare its category from §3.

What about plugins that span categories?#

Some plugins are easier to think about as “things” than as patterns. A “personal assistant skill bundle” might add new tools, customise channel behaviour, AND register memory hooks. We catalog those by their primary category in §4 with cross-references to the others.

What we are NOT going to claim#

The §3-vs-§4 split is editorial. Reasonable people might prefer “everything in one big catalog” or “each official category as its own top-level tab.” Our split is what we found cleanest for a learner. If it feels wrong to you, tell us and we’ll iterate.

Sources

See also