Skip to content
MCP server

记忆梦境引擎 · AI 的睡眠系统

Offline memory consolidation engine giving AI agents decay, scoring and a dream cycle

Pricing
Open source
Platforms
CLI, self-hosted, API
Version
1.0.1
Listing updated
today

Similar MCP servers

About 记忆梦境引擎 · AI 的睡眠系统

记忆梦境引擎 · AI 的睡眠系统 is an open-source MCP server for knowledge and memory. It is available on CLI, self-hosted, API, Linux, macOS and Windows.

The README describes memory-dream-engine as a pluggable pipeline that runs on a schedule (every 180 minutes in the author's Hermes Agent setup) and moves through collection, scoring, writing, optimization and content sprouting. Facts are pulled out with regular expressions instead of an LLM call, so the pipeline itself spends no model tokens.

Memory strength follows an Ebbinghaus-style curve with a 30-day half-life, where frequently accessed items are reinforced and low-strength items are archived. Signals are scored on three dimensions with weights that depend on signal type, so decisions, corrections and tool notes are not judged by one scale.

The project is a Python standard-library package with no third-party runtime dependencies, and it does not bind to a specific agent framework: the host supplies search, memory-write and usage functions as plain Python callables. It also ships as an MCP server over stdio or streamable HTTP, and its README says the server stores no caller data, with simulate_dream using a temporary database that is deleted after the run.

Tools this server provides (6)

Taken from the project documentation; the running server may differ.

extract_facts
Zero-LLM fact extraction using regular expressions.
score_signal
Three-dimensional dynamic scoring with a keep-or-discard verdict.
decay_report
Ebbinghaus decay report covering strength, reinforcement, archive status and curve.
simulate_dream
Runs a full five-stage dream cycle on supplied text using a temporary database.
list_signal_types
Lists signal types, weights and scoring keywords.
engine_info
Overview of capabilities, trade-offs against similar approaches and service boundaries.

What it does

Regex-based fact extraction
Extracts facts from Chinese and English text without calling a language model.
Ebbinghaus memory decay
Weakens memories over a 30-day half-life, reinforces frequently accessed items and archives weak ones.
Three-dimensional signal scoring
Scores decision, correction and tool signals with type-specific weighting.
Five-stage dream cycle
Runs collection, scoring, writing, optimization and sprouting in sequence on a schedule.
Content sprouting prompts
Suggests drafting a post or article once enough related memories accumulate.
Framework-agnostic adapters
Plugs into any agent through three user-supplied search, memory and usage functions.

Good for

  • Keeping an agent's long-term memory from filling up with stale notes
  • Scoring whether a conversation fact deserves to be saved to persistent memory
  • Testing a memory consolidation cycle on sample text before deploying it
  • Turning accumulated agent notes into draft content ideas

Pricing

Open source

Open source under the MIT license; the README lists no paid plan.

Prices change. Check savantcat.cn before you buy.

记忆梦境引擎 · AI 的睡眠系统 FAQ

Does the engine call an LLM?

No. The README says fact extraction is pure regular expression matching, so it does not use model tokens.

What dependencies does it need?

The README states it has zero third-party runtime dependencies and uses only the Python standard library; pytest is needed only for running tests.

Does the MCP server store my data?

The README says the server is pure computation and read-only, keeps no caller data, and that simulate_dream uses a temporary database that is deleted after the run.

Is 记忆梦境引擎 · AI 的睡眠系统 free?

Yes. 记忆梦境引擎 · AI 的睡眠系统 is open source, so you can use and self-host it for free.

What are the best alternatives to 记忆梦境引擎 · AI 的睡眠系统?

The closest alternatives to 记忆梦境引擎 · AI 的睡眠系统 are Dreamd, agentmem, Total Agent Memory, Holographic Memory and Brain Memory.

Where this comes from

  • Description: written from the project's README, the project's website and its MCP Registry entry.

How AIPoweredBot checks servers