arXiv cs.CL· Markus Fischer, Jens Meiler, Matthias Elgeti·· 12 小时前AI 评分58
MuninAI：面向独立研究组的自托管智能体框架
MuninAI: A Self-Hosted Agentic Framework for Independent Research Groups
AI 导读
MuninAI 是一个开源的自托管科研助手框架，索引研究组自有文献库与引用图，并用带阅读、搜索和计算工具的智能体编排层封装可本地部署的开源权重模型。在 199 个配对文献问题上，该编排层把基座模型准确率从 0.387 提升到 0.874（+0.487，95% CI [0.407, 0.568]），而朴素检索后回答在每个测试基座上都低于无辅助模型。
正文 · 原文
Abstract:Research groups that want a language model grounded in their own literature face a choice between hosted services that receive their unpublished work and self-hosted deployments that keep the data local but give the model no access to the corpus. Munin is an open-source framework for a self-hosted scientific assistant that indexes a group's corpus and its citation graph, and wraps whichever open-weight model the group can serve in an agent harness with specialised reading, search, and computation tools. We claim that the harness, rather than the specific model, is what makes the system useful on domain literature, and we ship the evaluation suite that tests this claim with the software.
On 199 paired literature questions, the harness more than doubles the accuracy of the base model, from 0.387 to 0.874 (+0.487, 95% CI [0.407, 0.568]), while naive retrieve-then-answer scores below the unaided model on every backbone tested. Under the harness, roughly twice as many answer claims trace to a passage the model read itself as with fixed top-5 retrieval. When the source paper for a question is removed from the corpus, the system abstains on about nine in ten of the questions it had answered correctly on the Qwen backbones and eight in ten on gpt-oss-20b, and it produces no fabricated local citation across 300 fabricated references.
The pattern holds on three backbones from two developers, including a 20-billion-parameter model that runs on a single consumer-grade GPU. With all outbound network access disabled, both Qwen backbones still exceed the published PaperQA2 accuracy. The evaluation suite runs as a single command, so an adopting group can re-run it on its own corpus and model.
| Comments: | 37 pages, 3 figures, 20 tables. Code, evaluation suite and scorecards: this https URL (archived release v1.0.0: this https URL) |
| Subjects: | Computation and Language (cs.CL) |
| ACM classes: | H.3.3; I.2.7 |
| Cite as: | arXiv:2610.23480 [cs.CL] |
| (or arXiv:2610.23480v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.23480 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Markus Fischer [view email]
[v1]
Fri, 16 Oct 2026 13:56:37 UTC (180 KB)
来源：arXiv cs.CL · arxiv.org