Claude remembers your code.
Even when you don't.
Mneme is a local-first persistent memory layer for AI coding assistants. It survives Claude's context compaction, remembers your codebase across sessions, and runs entirely on your laptop — no cloud dependencies, no API keys.
One line. Any platform.
Native macOS, Linux, Windows. Every install downloads directly from dist.stryxlabs.com, verifies SHA-256, and registers the MCP server.
$ curl -fsSL https://mneme.stryxlabs.com/install.sh | bash
$ curl -fsSL https://mneme.stryxlabs.com/install.sh | bash
PS> irm https://mneme.stryxlabs.com/install.ps1 | iex
$ mneme daemon start
$ mneme build .
$ mneme recall "auth handler"
$ mneme doctor
49 tools. Zero cloud.
Every query the AI needs — recall, blast radius, call graph, audit suite — local, deterministic, audited.
Core query
- find_references
- call_graph
- blast_radius
- recall_concept
- god_nodes
- dependency_chain
- architecture_overview
- smart_questions
Code analysis
- detect_patterns
- find_duplicates
- convention_check
- security_audit
- performance_scan
- drift_detection
File & symbol
- search_files
- read_file
- list_symbols
- get_symbol_info
- resolve_path
Memory & history
- step_resume
- step_verify
- recall_decision
- history_search
Workspace
- audit_corpus
- project_health
- storage_stats
- rebuild_shards
Multimodal
- ingest_pdf
- extract_images
- transcribe_audio
- process_video
Configuration
- get_config
- set_config
- export_config
- validate_rules
Admin
- health
- metrics
- doctor
- snapshot
- restore
- logs
- status
- version
What runs on your machine.
AI host → MCP wire → Mneme daemon → local store. Everything sits on your disk. Nothing routes to a third-party cloud.
One round-trip: blast_radius("handleLogin") takes ~20ms p95. Zero network calls.
Faster. Cheaper. Done.
Against 4 code-graph MCPs on a 50K+ LOC, 400+ file workspace.
| Metric | Mneme | tree-sitter | CRG | graphify |
|---|---|---|---|---|
| Total wall time | 411 s | 734 s | 1,111 s | 478 s |
| Cost per answer | $0.97 | $1.38 | $1.20* | $0.93 |
| Index build time | 23 s | per-query | 41 s | 13 s |
| Features shipped | 7 of 7 | 0 of 7 | 0 of 7 | 0 of 7 |
Annualized savings, 100 developers @ 50 queries/day: $512,500/yr vs tree-sitter · $287,500/yr vs CRG.
27 languages. One graph.
Tree-sitter grammars across 27 languages, with real symbol resolvers for Rust, TypeScript, and Python (more coming).
Stop grepping. Start recalling.
One line installs it. Your AI starts answering harder questions tomorrow.
Mneme · v0.4.1 Genesis · A work of Stryx Labs · Anish Trivedi & Kruti Trivedi · MMXXVI · License