Memory for Coding Agents

The Problem

Your coding agent forgets everything between sessions. Yesterday’s architecture discussion, the library you agreed on, the CSV you analyzed together — gone. Every session restarts from zero, and long-lived projects suffer most.

What MemoryLake Changes

The memorylake-openclaw plugin gives your agent three automatic behaviors, plus explicit tools:

Behavior	What happens
Auto-Recall	The agent is instructed to retrieve relevant memories and document context before answering each of your messages.
Auto-Capture	After each turn, the conversation is analyzed server-side — durable facts are stored, updated, or merged (deduplicated, incremental).
Auto-Upload	Files you share in chat are uploaded to your project in the background, becoming searchable documents.

The agent also gets explicit tools: retrieve_context (one call searches memories and documents, and surfaces memory conflicts), memory_store, memory_list, memory_forget, document_download, plus optional advanced_web_search and open_data_search for curated external datasets (arXiv, PubMed, SEC, FRED, USPTO, and more).

Setup

Get your key and project id

In the console, create an API key (API Keys in the sidebar) and note the project you want the agent to use. The console’s Integrations → OpenClaw page generates the exact commands below pre-filled.

Install the plugin

One-liner from the console (recommended — it configures everything):

curl -fsSL https://app.memorylake.ai/memorylake-openclaw/install.sh | bash -s -- \
  --project-id <your-project-id> --api-key sk-... --host https://app.memorylake.ai

Or manually:

openclaw plugins install memorylake-openclaw
openclaw gateway restart

then add to ~/.openclaw/openclaw.json:

{
  "plugins": {
    "entries": {
      "memorylake-openclaw": {
        "enabled": true,
        "config": {
          "apiKey": "sk-...",          // or "${MEMORYLAKE_API_KEY}"
          "projectId": "<project-id>",
          "host": "https://app.memorylake.ai"
        }
      }
    }
  }
}

The install script sets tools.profile to "full" — the plugin’s tools need it. A Windows PowerShell installer (install.ps1) and QClaw (~/.qclaw/openclaw.json) are supported the same way. Configuration hot-reloads: changing the key, project, or host takes effect without restarting the gateway.

Verify

Start a session and tell the agent something memorable, e.g. “Remember: we deploy on Fridays only.” End the session, start a new one, and ask “When do we deploy?” — the agent recalls it via retrieve_context.

Power Moves

One agent, many clients — without memory bleed

Create a project per client, then drop a per-directory override in each working folder:

// {workspace}/.memorylake/config.json
{ "projectId": "<client-a-project-id>" }

Sessions in that directory read and write only that client’s memory. The bundled agent-memorylake-config skill sets this up conversationally.

Bulk-load context

Feed the agent’s memory directly from the command line — single files, directories, or archives (zip/tar/7z/rar auto-extracted):

openclaw memorylake upload ./docs-archive.zip
openclaw memorylake search "authentication flow" --limit 10
openclaw memorylake stats

Migrate existing history

The bundled migrate-memories-to-memorylake skill imports memories and conversation history from your agent’s existing session files, so you don’t start from zero.

Controlled web access

advanced_web_search is registered as an optional tool (off unless you allow it) and respects plugin-level domain allowlists/blocklists — useful for enterprise environments that restrict agent browsing.

Use MemoryLake models too

The separate memorylake-models-openclaw plugin registers MemoryLake’s Model Router as a model provider — one sk-… key for both memory and models:

openclaw plugins install memorylake-models-openclaw
openclaw memorylake setup   # interactive: validates key, picks default model

What This Feels Like Day to Day

“Continue the refactor” — the agent recalls the module layout you agreed on two sessions ago.
You drag a stack trace file into chat — it’s auto-uploaded; next week the agent can still cite it.
“Why did we pick Postgres over MySQL?” — answered from the decision memory captured when you made the call, with its source trace.

Privacy & Telemetry

The plugin sends data only to your configured MemoryLake host. Usage telemetry is opt-in (disabled unless you explicitly set a telemetry key).

Getting started

Scenarios

Memory Router

Model Router

MemoryLake

Team collaboration

Memory for Coding Agents

The Problem

What MemoryLake Changes

Setup

Power Moves

One agent, many clients — without memory bleed

Bulk-load context

Migrate existing history

Controlled web access

Use MemoryLake models too

What This Feels Like Day to Day

Privacy & Telemetry

Go Deeper

Integrations Overview

Memories Guide

​The Problem

​What MemoryLake Changes

​Setup

​Power Moves

​One agent, many clients — without memory bleed

​Bulk-load context

​Migrate existing history

​Controlled web access

​Use MemoryLake models too

​What This Feels Like Day to Day

​Privacy & Telemetry

​Go Deeper

Integrations Overview

Memories Guide

The Problem

What MemoryLake Changes

Setup

Power Moves

One agent, many clients — without memory bleed

Bulk-load context

Migrate existing history

Controlled web access

Use MemoryLake models too

What This Feels Like Day to Day

Privacy & Telemetry

Go Deeper