Slowave

A living local memory layer across your AI tools.

Install once. Your AI tools share a persistent local memory across sessions and clients.

Slowave continuously adapts to your work — capturing decisions, preferences, and context over time.

  • Latent-space evolving memory, not yet another LLM summarizer or static vector store.
  • Lifelong learning through consolidation and decay.
  • No LLM API key required. €0 token cost.
  • 100% local. No data leaves your machine.

PyPI Python PyPI Status License: AGPL-3.0-or-later

How it feels

You work daily with your AI tools:

  • Day 1 — cold start: Slowave bootstraps memory from existing markdown knowledge, initializing the embedding-based memory state.
  • Week 1 — emerging patterns: new interactions begin reinforcing relevant signals, forming stable associations.
  • Month 1 — context consolidates: frequently reinforced information becomes consistently retrievable, low-signal data fades.

How Day 1 would look like with Slowave:

Demo

Multiple AI clients continuously build and reuse the same evolving memory over time:

  • no markdown management
  • no static RAG
  • no LLM extra calls

What you gain over time

Slowave becomes more useful the more you use it.

  • Clarity — your AI understands you without repeated explanation
  • Continuity — pick up projects where you left off
  • Consistency — keep your context across AI tools
  • Retention — retain decisions, patterns, and preferences over time
  • Focus — spend time creating instead of managing context

Slowave does not just store information — it compounds it into usable context.

The result is a continuous working context that follows you across tools and time.

Why Slowave is different

Slowave is a brain-inspired architecture.

Principle

The human brain does not need language to remember: experiences are encoded, replayed during sleep, abstracted into patterns, and strengthened or forgotten.

Language acts as the interface, not the storage medium.

Slowave mirrors this separation: the language model is a client of memory, not the memory system itself. Memory activity — consolidation, ranking, decay, retrieval — operates over embeddings rather than rewritten text.

Biological mapping

Neuroscience describes human memory as two complementary learning systems:

  • the hippocampus learns fast — it captures individual experiences and preserves them as distinct episodes;
  • the neocortex learns slowly — it extracts regularities across many experiences and transforms them into stable, general knowledge.

Neither system alone is sufficient. Fast learning without abstraction creates a collection of isolated events; slow learning without episodic grounding loses the experiences that shape knowledge.

Slowave mirrors this division:

Human brain Slowave What it does
Hippocampus Episodic layer Captures individual experiences as they occur
Neocortex Schema layer Extracts stable, abstract knowledge from repeated experiences into typed claims: decisions, preferences, constraints, conventions

Memory systems are supported by consolidation processes:

Human brain process Slowave What it does
Memory consolidation Offline consolidation Replays and groups episodes into prototypes, strengthens useful associations, and builds the memory graph without requiring an LLM

Three additional mechanisms emerge from this architecture:

  • Spreading activation. Recall propagates through the prototype association graph — a partial cue can recover related memories, similar to how one fragment of an experience can trigger a broader recollection.
  • Hebbian reinforcement. Memories that repeatedly prove useful become stronger and easier to retrieve; unused memories gradually decay. Forgetting improves signal-to-noise ratio.
  • Reconsolidation. Retrieval reopens memories for modification. Feedback — useful, stale, or incorrect — updates the memory state. Memory is dynamic, not an append-only log.

Design rationale

Architecture

Installation

Global setup

Install Slowave and configure every detected client in one go:

pipx install slowave

# or

brew tap mrsalty/slowave https://github.com/mrsalty/slowave
brew install slowave

Then wire everything up:

slowave setup --dry-run   # preview what will change
slowave setup             # apply: MCP configs, lifecycle instructions, hooks, services
slowave doctor            # verify: daemon health, client detection

slowave setup is idempotent and safe to run multiple times. The HTTP MCP daemon and background consolidation worker start automatically as system services.

Claude Desktop and Cursor require one manual paste after setup because their instruction surfaces cannot be modified programmatically. slowave setup prints the exact text and path.

Per-client setup

To configure a single client, or to find client-specific details:

Client Integration doc
Claude Code integrations/claude-code/README.md
Claude Desktop ¹ integrations/claude-desktop/README.md
Cline integrations/cline/README.md
Cursor ¹ integrations/cursor/README.md
OpenCode integrations/opencode/README.md
Windsurf integrations/windsurf/README.md
Codex integrations/codex/README.md

¹ requires one manual paste after setup

See the complete install & setup reference: docs/install.md

Storage

The default embedding model downloads from Hugging Face on first use (~45 MB, cached locally). Subsequent runs work offline.

Memory is stored in a local SQLite database at ~/.slowave/slowave.db — fully inspectable, never leaves your machine. Not encrypted by default; protect sensitive data with OS permissions or full-disk encryption.

Dashboard

Monitor Slowave’s health, incoming events, and memory consolidation in real time.

dashboard.png

Drill down from consolidated schemas to the underlying episodes, sessions, and raw events.

dashboard.png

Explore the evolving memory graph as Slowave forms connections across experiences.

dashboard_graph.png

Supported clients

Work in progress — suggest more integrations or report broken ones with setup details.

✅ = manually verified · ⬜ = pending verification

Client macOS Linux Windows Setup
Claude Code ✅ ✅ ✅ slowave setup --client claude-code
Cline ✅ ✅ ✅ slowave setup --client cline
Cursor ✅ ✅ ⬜ slowave setup --client cursor ¹
Windsurf (Devin) ✅ ✅ ⬜ slowave setup --client windsurf
Claude Desktop ✅ ✅ ✅ slowave setup --client claude-desktop ¹
OpenCode ✅ ✅ ✅ slowave setup --client opencode
Codex ✅ ✅ ✅ slowave setup --client codex
All the above slowave setup

¹ requires one manual paste after setup

Benchmarks

All runs: zero LLM calls, fully local, no API key. Two scorers reported: keyword-overlap (free, always computed — measures whether the right tokens appear in retrieved context) and LLM-judge (semantic grading via deepseek-v4-flash — comparable to Mem0/Zep's published numbers when they use the same judge model). Results have not yet been independently reproduced.

Benchmark n Keyword LLM-Judge What it tests
DMR 500 99.0% — Wikipedia-page factual recall
LongMemEval 500 87.8% 55.8% Multi-session facts, updates, preferences, temporal reasoning
LoCoMo 1,986 85.75% 69.29% Cross-session conversational recall across 10 real dialogues
StaleMemory 1,200 97% — Detecting when a stored preference silently changes

Full methodology and per-category breakdowns: docs/benchmarks.md.

Honest limits

  • It recalls stored information; it does not infer missing preferences.
  • It retrieves relevant memories; it does not perform reasoning.
  • Contradiction handling is heuristic and may not always resolve conflicts correctly.
  • Memory quality depends on the quality and consistency of prior interactions.

See: docs/limitations.md

What it is not

Slowave is not:

  • an agent framework
  • a reasoning system
  • a prompt manager
  • a markdown-based memory store
  • a vector database wrapper

The AI client remains responsible for planning, reasoning, and execution.

Slowave provides relevant, persistent, evolving context injection based on prior interactions.

Documentation

Contributing

Slowave is open source under the AGPL-3.0-or-later license.

Contributions are welcome, especially in:

  • client integrations
  • recall quality improvements
  • evaluation datasets
  • performance optimization

See CONTRIBUTING.md before submitting a pull request.

Project links

Key dates

PyPI data
Data sourced directly from PyPI's database.
  • Released:
Newer release available (0.18.1)

1 maintainer

PyPI data
Data sourced directly from PyPI's database.
Avatar for mrsalty from gravatar.com mrsalty

Credits

Author: mrsalty

License

AGPL-3.0-or-later

Requires

Python >=3.11

Provides Extra

test http dev

Tags

memory agent ai mcp neuroscience rag llm

Classifiers

Report project as malware

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

slowave-0.16.2.tar.gz (422.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

slowave-0.16.2-py3-none-any.whl (439.6 kB view details)

Uploaded Python 3

File details

Details for the file slowave-0.16.2.tar.gz.

File metadata

  • Download URL: slowave-0.16.2.tar.gz
  • Upload date:
  • Size: 422.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.13

File hashes

Hashes for slowave-0.16.2.tar.gz
Algorithm Hash digest
SHA256 fa472f8302d8f0719d6270df9b2bca37bb04549d8cdf55a15c7ac1249a90c1bb
MD5 a69c14cc957981760488a8b0e1e1db9a
BLAKE2b-256 2a45103e244b19d3e6f695370c229fa228378edad7a45c05e359f7d2e43b9cef

See more details on using hashes here.

File details

Details for the file slowave-0.16.2-py3-none-any.whl.

File metadata

  • Download URL: slowave-0.16.2-py3-none-any.whl
  • Upload date:
  • Size: 439.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.13

File hashes

Hashes for slowave-0.16.2-py3-none-any.whl
Algorithm Hash digest
SHA256 69eadf09e2e1b1b8eba4a1d952bbfe5f5e1f499f3881c771db9c9c2a68ff4a50
MD5 a98add9855e99f5da473e89f0b8d73c1
BLAKE2b-256 6f945cdd53ecc4ee61fba5c31db560258b2494a5df057ac34f3171a42fa95361

See more details on using hashes here.

Release history Release notifications | RSS feed

0.18.1

2 files

0.18.0

2 files

0.17.5

2 files

0.17.4

2 files

0.17.3

2 files

0.17.2

2 files

0.17.1

2 files

0.17.0

2 files

0.16.4

2 files

0.16.3

2 files

This release

0.16.2 This release

2 files

0.16.1

2 files

0.16.0

2 files

0.15.6

2 files

0.15.5

2 files

0.15.4

2 files

0.15.3

2 files

0.15.2

2 files

0.15.1

2 files

0.14.4

2 files

0.14.3

2 files

0.14.2

2 files

0.14.1

2 files

0.14.0

2 files

0.13.1

2 files

0.13.0

2 files

0.12.1

2 files

0.12.0

2 files

0.11.1

2 files

0.11.0

2 files

0.10.0

2 files

0.9.4

2 files

0.9.3

2 files

0.9.2

2 files

0.9.1

2 files

0.9.0

2 files

0.8.2

2 files

0.8.1

2 files

0.8.0

2 files

0.7.0

2 files

0.6.1

2 files

0.6.0

2 files

0.5.11

2 files

0.5.10

2 files

0.5.9

2 files

0.5.8

2 files

0.5.7

2 files

0.5.6

2 files

0.5.4

2 files

0.5.3

2 files

0.5.2

2 files

0.5.1

2 files

0.4.9

2 files

0.4.8

2 files

0.4.7

2 files

0.4.6

2 files

0.4.5

2 files

0.4.4

2 files

0.4.2

2 files

0.4.1

2 files

0.4.0

2 files

0.3.0

2 files

0.2.0

2 files

0.1.18

2 files

0.1.17

2 files

0.1.16

2 files

0.1.15

2 files

0.1.14

2 files

0.1.13

2 files

0.1.12

2 files

0.1.11

2 files

0.1.10

2 files

0.1.9

2 files

0.1.8

2 files

0.1.7

2 files

0.1.6

2 files

0.1.5

2 files

0.1.3

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page