The context window resets
Every new Claude Code session starts without the decisions, dead ends, and project constraints from yesterday. You re-explain the same context, or the agent repeats the same mistake.
Memory for Claude Code
Persistent memory for Claude Code, without the cloud.
Awareness is a local-first memory layer for Claude Code. It stores decisions, workflows, pitfalls, and preferences as Markdown on your machine, injects workflow rules so Claude Code recalls context at session start, and makes zero LLM calls on retrieval.
Install locally
npx @awareness.market/setup --ide claude-codeNo account, no API key, no credit card. Setup detects Claude Code and writes the local daemon and workflow rules.
The problem
Claude Code starts every session from zero.
CLAUDE.md is instructions, not memory
CLAUDE.md is the right place for conventions and constraints. It is not a record of what happened: which approach failed, which customer objected, which task is still open.
Knowledge is trapped in old sessions
The most valuable context is often buried in a past conversation. Without structured recall, the agent cannot use it when the next decision depends on it.
How it works
Recall before work. Record after.
Initialize
awareness_init loads cross-session context at the start of a Claude Code session: knowledge cards, open tasks, active skills, agent profiles, and workflow rules.
Recall before work
awareness_recall runs hybrid semantic + full-text search across structured cards, sessions, and graph facts. Retrieval is local and makes zero LLM calls.
Record after change
awareness_record turns decisions, fixes, risks, and preferences into structured knowledge, with deduplication and conflict detection.
Awareness injects workflow rules into Claude Code's configuration, so the agent initializes memory at session start, recalls before work, and records decisions after every change. Retrieval runs on local SQLite with hybrid semantic + full-text search; it makes zero LLM calls.
What gets remembered
Knowledge, not a transcript.
- Decisions and the reasoning behind them
- Problem-solution pairs and failed approaches
- Project workflows and conventions
- Pitfalls, risks, and open questions
- Action items and task handoffs
- Reusable skills and agent roles
- Personal preferences and working style
- Cross-project knowledge shared across agents
Compare
Where Awareness fits.
Native CLAUDE.md + Auto Memory
Best for durable project instructions and simple preferences. Awareness adds structured episodic memory, cross-agent recall, and conflict detection.
claude-mem
A Claude Code-focused session capture plugin. Awareness is a cross-agent memory layer with the same local-first principle and a broader knowledge model.
Mem0
A managed memory platform with cloud and SDK options. Awareness is local-first by default, runs retrieval without LLM calls, and writes workflow rules into the IDE.
Read the comparisonFrequently asked questions
Is Awareness free for Claude Code?
Yes. Local-first memory is free forever: unlimited memory libraries, local SQLite, hybrid recall, graph memory, insight extraction, and conflict detection. There is no account or credit card required. Optional inference features in the marketplace use credits.
Does Awareness send my code or prompts to the cloud?
No, by default. Awareness stores memory as human-readable Markdown in your project .awareness directory, with a local SQLite index. Cloud sync is optional and only enabled if you choose it.
How is Awareness different from CLAUDE.md?
CLAUDE.md is for durable instructions and conventions. Awareness is for episodic and semantic memory: decisions, workflows, pitfalls, risks, action items, skills, and preferences learned across sessions. It writes workflow rules so Claude Code recalls before work and records after changes.
How is Awareness different from claude-mem?
claude-mem focuses on capturing and retrieving session context for Claude Code. Awareness is a cross-agent memory layer: the same memory can be used from Claude Code, Cursor, Codex, Windsurf, Cline, and other MCP clients, with structured knowledge cards and conflict detection.
Does Awareness work with Cursor too?
Yes. The same local memory layer works across Claude Code, Cursor, Codex, Windsurf, Cline, GitHub Copilot, Zed, JetBrains, and other MCP-compatible agents. Setup detects the IDE and writes the appropriate rules.
How do I verify the install?
Run npx @awareness.market/setup --ide claude-code, then open Claude Code and run /mcp. You should see awareness-memory in the MCP server list. Run awareness_init to load context for the session.
Give your agent a memory that compounds.
Local-first, no account, no credit card. Start in 30 seconds and keep the data on your machine.