AutoGineer
Generated banner for membukkit
Generated logo for membukkit

membukkit

Memory serverOfficial Memseek project

Created by Memseek

Published in memseekai/membukkit. Publisher page

22 GitHub stars

What it is

Long-term memory for LLM apps that shows its work: dated facts, as-of answers, supersession, and receipts on every answer. Library, CLI, local GUI, MCP.
— the description by Memseek

AutoGineer overview, from the brain’s own files

MemBukkit is a memory layer for LLM applications designed to solve the problem of persistent, explainable knowledge. It transforms raw conversations and documents into dated atomic facts, maintaining a dual-lane index of atomic facts and verbatim source turns. When information changes, the system marks old facts as superseded rather than deleting them, allowing agents to perform temporal queries (e.g., 'what was true in May?'). Every answer generated by the system includes detailed receipts, providing transparency into the evidence used, source references, and cost metrics. The architecture is built for performance and explainability, utilizing bucketed retrieval to minimize token usage while maintaining high accuracy in knowledge-intensive tasks.

How it works

Memory model: Vector store

AutoGineer overview, from the brain’s own files

The system operates through a write path that distills input into dated atomic facts and verbatim turns, storing them in a flat embedding index. During the read path, a question is processed with an as-of date, triggering a budgeted search across topic buckets to retrieve relevant evidence. The system then generates an answer accompanied by receipts that detail the specific facts used, their status, and their source pointers.

Folders
FolderRoleFiles
.githubagent-config3
docsdocs32
examplesexamples7
scriptsscripts5
srcsource-code194
teststests41
benchmarksfolder61
notebooksfolder4
uifolder23

Requirements

Apps: Docker

Runtimes: Python >=3.10

API keys (by name): ANTHROPIC_API_KEY, COMPAT_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY, TURBOPUFFER_API_KEY

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Google Antigravity, Kiro, OpenCode, Amp, Roo Code, Cline, JetBrains Junie, Devin (ex-Windsurf)

Set it up

  1. Not scanned by AutoGineerIts dependencies are installed by pip or uv when you run it and were not scanned by AutoGineer (6 declared). https://pypi.org/project/membukkit/

  2. Created by Memseek. Published in memseekai/membukkit. Licence: Apache License 2.0.

Usage

AutoGineer overview, from the brain’s own files

  • Python library integration for custom LLM applications
  • CLI tool for managing local memory stores
  • MCP server for connecting agents like Cursor or Claude Desktop to persistent memory
  • HTTP service for multi-tenant agent memory access
“How much is my rent?”
The agent returns the rent amount as of the specified date, citing the specific lease agreement or update note used as evidence.
“What did I decide about the migration?”
The agent retrieves the relevant dated facts from the memory store and provides a summary with a receipt showing the source conversation turn.

Documentation by Memseek

<p align="center"> <img src="https://raw.githubusercontent.com/memseekai/membukkit/main/logo.png" alt="MemBukkit" width="160"/> </p>

<h1 align="center">MemBukkit</h1>

<p align="center"> <a href="https://pypi.org/project/membukkit/"><img src="https://img.shields.io/pypi/v/membukkit?style=flat-square&color=3775a9&logo=pypi&logoColor=white" alt="PyPI"/></a> <a href="https://github.com/memseekai/membukkit/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/memseekai/membukkit/ci.yml?branch=main&style=flat-square&label=CI" alt="CI"/></a> <a href="https://github.com/memseekai/membukkit/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202.0-blue?style=flat-square" alt="License"/></a> <a href="https://www.python.org/downloads/"><img src="https://img.shields.io/badge/python-3.10%2B-3776ab?style=flat-square&logo=python&logoColor=white" alt="Python"/></a> <a href="https://github.com/memseekai/membukkit/blob/main/docs/index.md"><img src="https://img.shields.io/badge/docs-guide-0d9488?style=flat-square" alt="Docs"/></a> </p>

MemBukkit is long-term memory for LLM apps that shows its work. It turns conversations and documents into dated atomic facts. When something changes (new rent, new job), the old fact is superseded, not deleted, so you can ask "what was true in May?" and get May's answer. Every answer ships with receipts: the facts used, where they came from, what the query cost. It sets the state of the art on LongMemEval-S under the benchmark's official judge, 92.6%, while reading a small fraction of the tokens full-context reading pays. Python, CLI, local GUI, HTTP, and MCP, all over the same stores. Apache-2.0.

<p align="center"> <a href="https://github.com/memseekai/membukkit/blob/main/docs/index.md">Docs</a> · <a href="https://github.com/memseekai/membukkit/blob/main/docs/guide/install.md">Install</a> · <a href="https://github.com/memseekai/membukkit/blob/main/docs/guide/quickstart.md">Quickstart</a> · <a href="https://github.com/memseekai/membukkit/blob/main/docs/guide/agents.md">Agents</a> · <a href="https://github.com/memseekai/membukkit/blob/main/docs/guide/mcp.md">MCP</a> · <a href="https://github.com/memseekai/membukkit/blob/main/docs/guide/demos.md">Demos</a> · <a href="https://github.com/memseekai/membukkit/blob/main/docs/guide/benchmarks.md">Benchmarks</a> </p>

<p align="center"> <img src="https://raw.githubusercontent.com/memseekai/membukkit/main/docs/assets/demo-ask.gif" alt="Ask with receipts" width="720"/> </p>

Try it in one line

uvx --from "membukkit[all]" membukkit ui --demo personal-assistant

That opens the GUI on a preloaded demo store. No clone, no venv, and Node is never needed because the GUI ships prebuilt. With pip it's pip install "membukkit[all]". Docker works too:

docker run -p 127.0.0.1:8377:8377 -v membukkit-data:/data -e OPENAI_API_KEY ghcr.io/memseekai/membukkit

It runs fully local. Point it at Ollama and nothing leaves your machine:

membukkit ask "what did I decide about the migration?" --store notes --llm ollama:llama3.1

Prefer a hosted model? Paste a key in the GUI, or export OPENAI_API_KEY=sk-.... Either way, the first ask downloads the retrieval weights once and caches them.

Every install path, including a dev clone: Install guide.

Python API

from membukkit import Memory

mem = Memory.from_pretrained(llm="openai:gpt-4o-mini")
mem.add("I signed a lease, rent is $2100.", subject="alex", date="2024-01-10")
mem.add("Landlord raised rent to $2300.", subject="alex", date="2024-03-01")

r = mem.ask("How much is my rent?", as_of="2024-06-01")
print(r.answer)             # truth as of that date
print(r.est_reader_tokens)  # ~tokens the reader saw
print(r.evidence[0].status, r.evidence[0].source_ref)  # current|superseded + citation

add returns a write receipt (n_stored, superseded, status), so empty LLM extracts never look like success. The full CLI walkthrough of the same flow is in the Quickstart.

How it works

Long-term memory is an indexing problem before it is a reading problem. What the store commits to at write time decides what any reader can possibly see later. Everything in MemBukkit follows from that.

flowchart LR
  subgraph W["write path"]
    S["sessions & documents"] --> D["LLM distillation"]
    D --> F["dated atomic facts<br/>+ source pointers"]
    S --> V["verbatim turns"]
  end
  F --> X[("flat embedding index<br/>+ topic buckets")]
  V --> X
  subgraph R["read path"]
    Q["question + as-of date"] --> B["budgeted bucket routing"]
    B --> K["top-k dated evidence"]
    K --> A["answer + receipts"]
  end
  X --> B

Writing. An LLM distills each session or document into dated atomic facts like 2024-01-08: rent is 800€, and the verbatim source turns are stored right next to them. Every fact keeps a pointer to its source. Why both? Because an extractor decides at write time what will matter later, and it is sometimes wrong. The atomic lane gives the system dates and updates it can reason about. The verbatim lane keeps everything the extractor skipped. When a new fact contradicts an old one, the old one is marked superseded instead of deleted. That is what makes "what was true in May?" answerable.

Indexing. Both lanes go into one flat embedding index, optionally split into topic buckets with plain KMeans. No LLM-built graph, no ontology, no summary hierarchy. Structures like those bake their builder's assumptions into the index, and revising them means rebuilding everything. Here the index builds in seconds with zero LLM calls, and changing how you retrieve is a config change, not a migration.

Reading. A question opens the closest topic buckets until a scan budget is covered, then the reader answers from the top evidence: dated, source-linked lines, filtered to what was known at the as-of date. It reads a small slice of memory instead of everything, and it answers better because of it. See Benchmarks for how much better.

Receipts. Every answer reports what it used and what it cost: tokens, estimated dollars, scan fraction, opened buckets, and each evidence line's status with its source turn or clause. The receipts are not decoration. In our research, hiding exactly the buckets a receipt names destroys the answer, while hiding random ones changes nothing. The trace really is the evidence trail.

Optional: a BM25 lexical lane. Everything above is dense retrieval, which is the shipped method and the one every number here was measured with. If your corpus is full of exact strings that embeddings blur together, error codes, filenames, rare identifiers, you can turn on a BM25 lane that searches the whole bank by term overlap and adds its hits to the routed pool before ranking. It is off by default and changes nothing until you ask for it:

from membukkit import Memory
from membukkit.config import RetrievalConfig

mem = Memory.from_pretrained(retrieval=RetrievalConfig(lexical_lane=True))

Needs pip install "membukkit[bm25]". How it fuses and what it costs: Library API.

Chosen by answer quality, not retrieval metrics

The research behind MemBukkit kept producing the same surprise: better retrieval did not mean better answers. Reading everything scored far below reading a routed slice. A stronger reranker did not beat plain cosine order. Extraction-heavy designs lost evidence that a flat two-lane index kept. So every retrieval policy that ships here was selected by one test: did the final answer get better?

That is also why nothing depends on our fine-tuned models. Swap in off-the-shelf weights, or an all open-weights stack, and the results hold. You are adopting an index design, not a checkpoint.

The numbers behind each of these claims are in Benchmarks below. The full mechanism is in Method, and a research paper is under review. For how MemBukkit compares to file notes, plain vector RAG, and temporal graphs as categories, see When to use.

Benchmarks

Every score here is graded by each benchmark's official judge, and every one is a frozen recipe that pins the reader, distiller, judge, encoder, and distillation cache. One command reruns it, and --check verifies your result against the expected band.

That first part is doing more work than it looks. Higher LongMemEval numbers exist, and they are graded by their own authors, in one case by the same model that wrote the answers. Under the official gpt-4o judge the field is MemBukkit 92.6, then Supermemory 85.2, then Zep 71.2. See who judges what.

Benchmark (what it stresses)StackScoreReproduce
LongMemEval-S (knowledge updates, temporal reasoning across sessions)gpt-5.4 reader, official gpt-4o judge92.6%membukkit bench --repro longmemeval-gpt54
LongMemEval-Sgpt-4o-mini reader82.0% (95% CI 78.6–85.4)membukkit bench --repro longmemeval-gpt4o-mini
LongMemEval-Sall open weights: gemma-4-26b reader + distiller88.8%membukkit bench --repro longmemeval-gemma
LoCoMo (Mem0's protocol and judge, zero retuning)gpt-4o-mini87.5%membukkit bench --repro locomo-mem0
BEAM (100K / 1M / 10M-token haystacks, official judge)gemma-4-26b0.535 / 0.498 / 0.447membukkit bench --repro beam-100k-gemma

What the research found, on LongMemEval-S (500 questions, official gpt-4o judge, paired comparisons sharing reader, judge, and ingestion):

  • Full-context reading scores 56.4% against MemBukkit's 82.0% with the same reader and judge, a +25.6 point gap (paired 95% CI [20.8, 30.4]), while MemBukkit reads ~3.2k tokens per question instead of ~100k.

  • Excluding the buckets an answer's receipt names collapses accuracy from 80.0% to 1.3%. Excluding a matched random set leaves it at 82.3%.

  • Given identical ingestion, extraction-only Mem0 scores 21.4% on questions answered in the assistant's own replies, where the verbatim lane scores 92.9%. Lane ablation: verbatim-only 75.4%, atomic-only 58.0%, both 82.8%.

  • Dropping the cross-encoder entirely and ranking the same opened region by plain cosine (83.4%, --rerank-select none) statistically ties the shipped hybrid config (82.0%, paired p=0.40). Swapping the fine-tuned encoder for an off-the-shelf one moves the score by just −0.4 (p=0.81). The reranker is tail insurance, not the headline.

Single-pass ask, no agentic re-query loops, official judges. LLM readers and judges are stochastic, so scores reproduce as bands, not bit-exact values. Judges, costs, and per-category breakdowns: Benchmarks guide.

Retrieval, measured directly

Answer quality is one thing, retrieval is another, so there is a second suite that grades retrieval alone with no LLM and no judge in the loop: HotpotQA multi-hop, 1,000 questions, scored against qmd by the same metric functions over the same candidate sets.

A multi-hop question needs two documents, and one of two is a miss, not half a success. On finding both, MemBukkit's document-retrieval path reaches 87.4% against qmd's best backend at 86.1% while answering in 47ms against 1,222ms. A heavier configuration, swapping in a 0.6B encoder and adding query decomposition driven by a local 1.7B model, reaches 90.8%. On the bridge questions that actually require two hops the gap is 88.4% against 82.3%. qmd stays a little ahead at putting one right document in first place, 93.2% against 92.1%, a difference inside sampling noise.

The same no-LLM treatment on conversational memory: session-level Recall@5 on the original LongMemEval-S split, the metric gbrain and MemPalace publish against. With gbrain's own embedding model (text-embedding-3-large) under MemBukkit's turn-level indexing and no reranker at all, 99.8% (499/500) against gbrain-hybrid's 97.6% and MemPalace's 96.6%, same dataset, same K, same n, no LLM in the loop, none of our fine-tuned weights involved. The as-shipped production cut with the same encoder scores 99.2%; with our 110M local encoder and zero API calls, 96.4%. Per-question rankings, a training-overlap disclosure for the shipped reranker, and the commands that regenerate every row are in benchmarks/longmemeval_session_recall/.

Read it with the setup, which is in the benchmark notes: those rows measure the document-retrieval configuration of the stack rather than the chunked path membukkit search runs today, the candidate sets are small enough that both systems saturate at k=10, and qmd's lexical backend scored zero on these long natural-language queries in a way that looks like a configuration artifact rather than a real result.

Known limits, so nothing surprises you later: distillation is an LLM call per session at write time (cached), so the cost moves rather than vanishing. Updates supersede rather than overwrite, so the old fact stays stored, which is what makes as-of queries possible; when you want a fact gone for real, delete it. And the distiller bounds the atomic lane: small local models extract weaker facts, and the verbatim lane limits the damage.

Bring your data

membukkit ui                                     # create a store → Ingest (drag & drop)
membukkit ingest ~/Downloads/WhatsAppChat.txt --store me   # or from the CLI
Drop thisHow you get it
WhatsApp .txtChat → Export → without media
ChatGPT / Claude ZIPSettings → Export data (or conversations.json)
PDFs / CRM CSV / notesHubSpot deals export, contracts, Notion MD zip, Obsidian folder, ./docs

Ask the same question at two dates and compare the receipts. Full steps per source: Bring your own.

MCP

A thin stdio server gives Cursor / Claude Desktop three tools over the same stores: memory_add, memory_search, memory_ask.

membukkit mcp --store notes

Client config and example prompts: MCP guide.

Surfaces

PieceRole
Memoryadd / search / ask / delete + write & ask receipts
CLI / GUIingest files, demos, explainability lab
Local v1 HTTP/api/v1/{store}/… for agents on disk stores
MCPCursor / Claude tools over the same stores

MemorySystem remains the full pipeline API. See Library API and Method.

Contributing & license

Contributions welcome, see CONTRIBUTING.md. Licensed under Apache-2.0.

LICENSE

Shown as plain text.


                                 Apache License
                           Version 2.0, January 2004
                        http://www.apache.org/licenses/

   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION

   1. Definitions.

      "License" shall mean the terms and conditions for use, reproduction,
      and distribution as defined by Sections 1 through 9 of this document.

      "Licensor" shall mean the copyright owner or entity authorized by
      the copyright owner that is granting the License.

      "Legal Entity" shall mean the union of the acting entity and all
      other entities that control, are controlled by, or are under common
      control with that entity. For the purposes of this definition,
      "control" means (i) the power, direct or indirect, to cause the
      direction or management of such entity, whether by contract or
      otherwise, or (ii) ownership of fifty percent (50%) or more of the
      outstanding shares, or (iii) beneficial ownership of such entity.

      "You" (or "Your") shall mean an individual or Legal Entity
      exercising permissions granted by this License.

      "Source" form shall mean the preferred form for making modifications,
      including but not limited to software source code, documentation
      source, and configuration files.

      "Object" form shall mean any form resulting from mechanical
      transformation or translation of a Source form, including but
      not limited to compiled object code, generated documentation,
      and conversions to other media types.

      "Work" shall mean the work of authorship, whether in Source or
      Object form, made available under the License, as indicated by a
      copyright notice that is included in or attached to the work
      (an example is provided in the Appendix below).

      "Derivative Works" shall mean any work, whether in Source or Object
      form, that is based on (or derived from) the Work and for which the
      editorial revisions, annotations, elaborations, or other modifications
      represent, as a whole, an original work of authorship. For the purposes
      of this License, Derivative Works shall not include works that remain
      separable from, or merely link (or bind by name) to the interfaces of,
      the Work and Derivative Works thereof.

      "Contribution" shall mean any work of authorship, including
      the original version of the Work and any modifications or additions
      to that Work or Derivative Works thereof, that is intentionally
      submitted to Licensor for inclusion in the Work by the copyright owner
      or by an individual or Legal Entity authorized to submit on behalf of
      the copyright owner. For the purposes of this definition, "submitted"
      means any form of electronic, verbal, or written communication sent
      to the Licensor or its representatives, including but not limited to
      communication on electronic mailing lists, source code control systems,
      and issue tracking systems that are managed by, or on behalf of, the
      Licensor for the purpose of discussing and improving the Work, but
      excluding communication that is conspicuously marked or otherwise
      designated in writing by the copyright owner as "Not a Contribution."

      "Contributor" shall mean Licensor and any individual or Legal Entity
      on behalf of whom a Contribution has been received by Licensor and
      subsequently incorporated within the Work.

   2. Grant of Copyright License. Subject to the terms and conditions of
      this License, each Contributor hereby grants to You a perpetual,
      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
      copyright license to reproduce, prepare Derivative Works of,
      publicly display, publicly perform, sublicense, and distribute the
      Work and such Derivative Works in Source or Object form.

   3. Grant of Patent License. Subject to the terms and conditions of
      this License, each Contributor hereby grants to You a perpetual,
      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
      (except as stated in this section) patent license to make, have made,
      use, offer to sell, sell, import, and otherwise transfer the Work,
      where such license applies only to those patent claims licensable
      by such Contributor that are necessarily infringed by their
      Contribution(s) alone or by combination of their Contribution(s)
      with the Work to which such Contribution(s) was submitted. If You
      institute patent litigation against any entity (including a
      cross-claim or counterclaim in a lawsuit) alleging that the Work
      or a Contribution incorporated within the Work constitutes direct
      or contributory patent infringement, then any patent licenses
      granted to You under this License for that Work shall terminate
      as of the date such litigation is filed.

   4. Redistribution. You may reproduce and distribute copies of the
      Work or Derivative Works thereof in any medium, with or without
      modifications, and in Source or Object form, provided that You
      meet the following conditions:

      (a) You must give any other recipients of the Work or
          Derivative Works a copy of this License; and

      (b) You must cause any modified files to carry prominent notices
          stating that You changed the files; and

      (c) You must retain, in the Source form of any Derivative Works
          that You distribute, all copyright, patent, trademark, and
          attribution notices from the Source form of the Work,
          excluding those notices that do not pertain to any part of
          the Derivative Works; and

      (d) If the Work includes a "NOTICE" text file as part of its
          distribution, then any Derivative Works that You distribute must
          include a readable copy of the attribution notices contained
          within such NOTICE file, excluding those notices that do not
          pertain to any part of the Derivative Works, in at least one
          of the following places: within a NOTICE text file distributed
          as part of the Derivative Works; within the Source form or
          documentation, if provided along with the Derivative Works; or,
          within a display generated by the Derivative Works, if and
          wherever such third-party notices normally appear. The contents
          of the NOTICE file are for informational purposes only and
          do not modify the License. You may add Your own attribution
          notices within Derivative Works that You distribute, alongside
          or as an addendum to the NOTICE text from the Work, provided
          that such additional attribution notices cannot be construed
          as modifying the License.

      You may add Your own copyright statement to Your modifications and
      may provide additional or different license terms and conditions
      for use, reproduction, or distribution of Your modifications, or
      for any such Derivative Works as a whole, provided Your use,
      reproduction, and distribution of the Work otherwise complies with
      the conditions stated in this License.

   5. Submission of Contributions. Unless You explicitly state otherwise,
      any Contribution intentionally submitted for inclusion in the Work
      by You to the Licensor shall be under the terms and conditions of
      this License, without any additional terms or conditions.
      Notwithstanding the above, nothing herein shall supersede or modify
      the terms of any separate license agreement you may have executed
      with Licensor regarding such Contributions.

   6. Trademarks. This License does not grant permission to use the trade
      names, trademarks, service marks, or product names of the Licensor,
      except as required for reasonable and customary use in describing the
      origin of the Work and reproducing the content of the NOTICE file.

   7. Disclaimer of Warranty. Unless required by applicable law or
      agreed to in writing, Licensor provides the Work (and each
      Contributor provides its Contributions) on an "AS IS" BASIS,
      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
      implied, including, without limitation, any warranties or conditions
      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
      PARTICULAR PURPOSE. You are solely responsible for determining the
      appropriateness of using or redistributing the Work and assume any
      risks associated with Your exercise of permissions under this License.

   8. Limitation of Liability. In no event and under no legal theory,
      whether in tort (including negligence), contract, or otherwise,
      unless required by applicable law (such as deliberate and grossly
      negligent acts) or agreed to in writing, shall any Contributor be
      liable to You for damages, including any direct, indirect, special,
      incidental, or consequential damages of any character arising as a
      result of this License or out of the use or inability to use the
      Work (including but not limited to damages for loss of goodwill,
      work stoppage, computer failure or malfunction, or any and all
      other commercial damages or losses), even if such Contributor
      has been advised of the possibility of such damages.

   9. Accepting Warranty or Additional Liability. While redistributing
      the Work or Derivative Works thereof, You may choose to offer,
      and charge a fee for, acceptance of support, warranty, indemnity,
      or other liability obligations and/or rights consistent with this
      License. However, in accepting such obligations, You may act only
      on Your own behalf and on Your sole responsibility, not on behalf
      of any other Contributor, and only if You agree to indemnify,
      defend, and hold each Contributor harmless for any liability
      incurred by, or claims asserted against, such Contributor by reason
      of your accepting any such warranty or additional liability.

   END OF TERMS AND CONDITIONS

   APPENDIX: How to apply the Apache License to your work.

      To apply the Apache License to your work, attach the following
      boilerplate notice, with the fields enclosed by brackets "[]"
      replaced with your own identifying information. (Don't include
      the brackets!)  The text should be enclosed in the appropriate
      comment syntax for the file format. We also recommend that a
      file or class name and description of purpose be included on the
      same "printed page" as the copyright notice for easier
      identification within third-party archives.

   Copyright [yyyy] [name of copyright owner]

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
   You may obtain a copy of the License at

       http://www.apache.org/licenses/LICENSE-2.0

   Unless required by applicable law or agreed to in writing, software
   distributed under the License is distributed on an "AS IS" BASIS,
   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
   See the License for the specific language governing permissions and
   limitations under the License.
384 files · 376 not served here

Advantages

  • Supports temporal reasoning by superseding old facts instead of deleting them, enabling accurate as-of queries. evidence: README.md
  • Provides full transparency with receipts that include evidence status, source references, and cost metrics for every answer. evidence: docs/index.md
  • Optimizes performance by using bucketed retrieval, reading only a small slice of memory rather than the full context. evidence: README.md

Limitations

  • Distillation is an LLM-dependent process that occurs at write time, incurring costs for each session or document ingested. evidence: README.md
  • The atomic lane is bounded by the quality of the distillation model; smaller local models may extract weaker facts. evidence: README.md

Security

Risk level: No risk found

This brain runs

This brain runs nothing: AutoGineer found no hooks, scripts, plugin code or MCP servers in its files.

Attribution and licence

Created by
Memseek
Declared authors
  • MemBukkit contributors (declared in pyproject.toml)
Publisher
Memseek (Official Memseek project)
Source
memseekai/membukkit · commit 66a3c82
Licence
Apache License 2.0 (files served here)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by Memseek.

Install from source

376 of its files are not served here, so there is no ZIP. memseekai/membukkit at 66a3c82