§ Connect a client
OpenClaw. Memory without the wiring.
OpenClaw is the open agent stack we ship first-party support for. Drop the plugin in and your agents stop forgetting. Recall happens before every prompt, decisions get captured as they're made, and you don't write a line of memory glue.
- Base URL
- platform.octamem.com
- Auth
- Bearer API key
- Primitives
- details / search / add

OctaMem × OpenClaw
First-party memory for the open agent stack
OctaMem installs as an OpenClaw plugin, so memory becomes part of the runtime rather than something you wire per agent. Recall runs automatically before each prompt and writes are captured as decisions are made — every record typed, scoped, and linked back to the source that produced it.
Install the plugin
Add the plugin
From your OpenClaw project root, install the OctaMem plugin and set your API key.
# In your OpenClaw project root:
openclaw plugin add octamem
openclaw config set octamem.api_key "oct_live_..."Configure recall and capture
Add the OctaMem block to your openclaw.toml. The defaults below are a sensible starting point, turn auto_recall and auto_capture on and tune the scope and intents from there.
# openclaw.toml
[plugins.octamem]
api_key_env = "OCTA_API_KEY"
default_scope = "myteam/agent-a"
# Auto-recall before every prompt.
auto_recall = true
recall_types = ["semantic", "episodic", "procedural"]
recall_limit = 8
# Auto-capture significant decisions back as memory.
auto_capture = true
capture_filter = "intent:decide,intent:commit"Run an agent
Start an OpenClaw agent the way you normally would. The plugin rewrites the system prompt with relevant memory before every model call, and writes captured decisions back to OctaMem in the background. You’ll see two new lines in your OpenClaw run log per turn: octamem.recall with the number of memories injected, and octamem.capture with the intent that triggered a write.
What you actually get
- Continuity. An agent that talked to a customer last Tuesday remembers their preferences this Tuesday, without you plumbing context through your own state machine.
- Decisions that stick.When an agent commits to an action, that fact lands in episodic memory automatically, with full provenance. No more “the agent said yes on Monday and no on Wednesday.”
- Scoped intents. The
capture_filterfield above takes OpenClaw intent tags.intent:decidefires when the agent makes a binding choice (approve, reject, route).intent:commitfires when the agent writes to a system of record (CRM, ticket, payment). Add or remove tags to control what becomes memory. - Audit you can hand to legal.Every read and write carries the agent’s OpenClaw identity, the run ID, and the tool call that triggered it.