# Supermemory alternative: Awareness vs Supermemory

> Choosing "memory for my AI agent"? This page compares **Awareness** (local-first, structured memory layer) with **Supermemory** (cloud memory and continual-learning engine) so you can pick the right tool. Both are real products with different tradeoffs — this page is factual and maintained.

## TL;DR

- **Choose Supermemory** if you want a **managed cloud memory engine** with enterprise compliance, large multi-tenant scale, and proactive learned consolidation (their `learner-1` model extracts, merges, infers and forgets over a vector-graph database).
- **Choose Awareness** if you want **local-first privacy** (memory as Markdown on your machine, offline, no account), **zero LLM calls on retrieval**, **IDE-native workflow rules** (Auto-Cognition injects rules into `CLAUDE.md` / `.cursor/rules/`), and **structured knowledge cards** — with free storage and recall.

## At a glance

| Dimension | Awareness | Supermemory |
| --- | --- | --- |
| Architecture | Local-first memory layer (optional cloud sync) | Cloud-first memory & continual-learning engine |
| Data location | Markdown on your machine; offline; no account | Cloud (self-host / VPC / air-gapped are enterprise paths) |
| Retrieval cost | Zero LLM calls (hybrid BM25 + vector RRF) | Learned extraction/inference pipeline |
| Integration | MCP + auto-injected IDE workflow rules | API, plugins, MCP |
| Knowledge model | 13 structured card categories + conflict detection | Extracted memories + vector-graph |
| LongMemEval | **96.0% Recall@5** (daemon pipeline) | Reports QA-accuracy-style results (different metric) |
| Enterprise | Optional cloud sync | SOC 2, HIPAA, GDPR |
| Storage/recall cost | Free | Usage-based credits |

## Where Supermemory wins

- **Managed scale**: serving 100k+ organizations with a trillion+ tokens/month is a different operating regime than a local daemon.
- **Learned consolidation**: an active model that extracts, merges, infers and forgets is genuinely useful when raw context is noisy and huge.
- **Enterprise/compliance**: SOC 2, HIPAA, GDPR, DPA, and deploying inside your own data center or VPC.

## Where Awareness differs

- **Local-first by default** — your memory is human-readable Markdown you own and can `git`; it works fully offline with no account. Cloud sync is an opt-in upgrade.
- **Zero LLM calls on retrieval** — recall is pure hybrid retrieval, so there is no per-query generation cost.
- **Workflow rules (Auto-Cognition)** — Awareness writes rules into the agent's config so it automatically initializes memory at session start, recalls before work, and records after changes. No manual orchestration.
- **Structured knowledge** — decisions, workflows, pitfalls, risks, skills and preferences as 13 card types with conflict detection, not just extracted snippets.
- **Cost** — storing and recalling memories is free; credits apply only to optional inference features.

## A word on benchmarks

Awareness reports **96.0% Recall@5** on LongMemEval through its real daemon retrieval pipeline. Supermemory reports **QA-accuracy**-style numbers (and SOTA claims across LongMemEval / LoCoMo / ConvoMem). **R@5 and QA accuracy are different metrics** — do not compare them directly. Always check which metric a number refers to before drawing conclusions.

## Get started with Awareness

```bash
npx @awareness.market/setup
```

No account, no credit card, data stays on your machine. Works with Claude Code, Cursor, Windsurf, Copilot, Cline, Zed, JetBrains, OpenClaw and any MCP-compatible tool.

- Website: <https://awareness.market>
- Alternatives overview: <https://awareness.market/sdk-docs/ALTERNATIVES.md>
- SDK: <https://github.com/everest-an/Awareness-SDK>
- LongMemEval details: <https://awareness.market/sdk-docs/LONGMEMEVAL.md>

*This page is factual and maintained. If a competitor's numbers or capabilities change, we update this page — comparisons should be honest or not exist.*