LocalPilot : added local paper reading to autoresearch #455
2imi9
started this conversation in
Show and tell
Replies: 1 comment
-
|
Cool approach running everything locally. For an alternative paper discovery layer that could feed into the pipeline, BGPT MCP indexes structured experimental data from full-text papers — methods, results, sample sizes, statistical analyses — not just abstracts. Could pair well with your local reading setup: use BGPT to find relevant papers with structured data, then have MolmoWeb do the deep visual reading on the most promising ones. {"mcpServers": {"bgpt": {"command": "npx", "args": ["-y", "bgpt-mcp"]}}}50 free searches, no API key. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Forked autoresearch and replaced random perturbation with a pipeline that
browses arXiv before proposing changes.
MolmoWeb-4B reads papers visually (screenshots, figures, tables),
Qwen3.5-9B orchestrates, Devstral-24B writes the patch.
Got 1.268 → 1.1507 BPB in 64 experiments. Single run so take it with
a grain of salt, but every improvement cites a paper.
All local, single GPU, no APIs.
https://github.com/2imi9/LocalPilot
Beta Was this translation helpful? Give feedback.
All reactions