AI coding assistant with persistent memory, powered by GitHub Copilot.
CoClaw adds a persistent memory layer to GitHub Copilot's LLM via the VS Code Language Model API. Instead of losing context between sessions, CoClaw stores and retrieves relevant memories — code conventions, user preferences, project patterns, past decisions — and injects them into LLM prompts automatically.
Automatic Memory Extraction — Facts, decisions, and preferences extracted from every conversation
Agentic Coding — Uses VS Code tools to read, edit, and search files autonomously with workspace-scoped safety
Model Switching — Click the status bar to switch between available Copilot models
Identity System — Customizable assistant persona (SOUL.json) and user preferences (USER.json)
Memory Browser — Webview panel to browse, edit, promote, and delete memory entries
LM Tools — Agent-mode tools for reading/writing memory mid-conversation
Import/Export — Backup and restore memories as JSON
Progress Indicators — Live status bar spinner and per-tool progress while working
Requirements
VS Code 1.93+
GitHub Copilot Chat extension
Usage
Chat
Type @CoClaw in the chat panel to start a conversation with memory-augmented responses.
Slash Commands
Command
Description
/memory
Show what CoClaw remembers
/distill
Distill daily logs into long-term memory
/clear
Clear today's session memory
/soul
Edit CoClaw's identity/behavior
Commands (Command Palette)
Command
Description
CoClaw: Select Model
Switch Copilot model via QuickPick
CoClaw: Browse Memory
Open memory browser webview
CoClaw: Clear Session Memory
Clear today's daily log
CoClaw: Edit Identity (SOUL)
Open SOUL.json in editor
CoClaw: Edit Profile (USER)
Open USER.json in editor
CoClaw: Distill to Long-Term Memory
AI-powered distillation of daily logs
CoClaw: Import Memories
Import memories from JSON file
CoClaw: Export Memories
Export all memories to JSON file
Settings
Setting
Default
Description
CoClaw.model.family
""
Preferred Copilot model family (workspace override)
CoClaw.memory.maxLongTermEntries
100
Max long-term memory entries
CoClaw.memory.dailyLogsRetentionDays
30
Days to retain daily logs
CoClaw.memory.autoExtract
true
Auto-extract facts from conversations
CoClaw.memory.tokenBudgetPercent
20
Max % of context window for memory injection
How Memory Works
During chat: CoClaw retrieves relevant memories and injects them into the system prompt
After each response: A secondary LM call extracts notable facts, preferences, and decisions
Daily logs: Raw extracted facts stored per-day as JSON
Long-term memory: Distilled, curated entries that persist indefinitely
Recall: Entries ranked by recency × importance × keyword_overlap and injected within token budget
Privacy
All memories are stored locally in VS Code's global storage. No data is sent to external services beyond the Copilot API calls. CoClaw enforces workspace boundaries — it will never read or write files outside your current workspace.
