-0.46% community-computer · fold visible_tip_oids into ticker pass +1.61% rad-model · chaining examples for stage+show — MAJOR improvement -90.89% wheresmy · cache SentenceTransformer at module level -64.28% icydream · add -flto to benchmark: cross-TU inlining wins +53.51% rad-model · re-run to measure API variability — new record -0.46% community-computer · fold visible_tip_oids into ticker pass +1.61% rad-model · chaining examples for stage+show — MAJOR improvement -90.89% wheresmy · cache SentenceTransformer at module level -64.28% icydream · add -flto to benchmark: cross-TU inlining wins +53.51% rad-model · re-run to measure API variability — new record
Human-centered · agent-scale

Collaborate at agentic scale.

Community Computer helps your team conduct experiments and sort through the noise to find what's important. Every result is signed and replicated amongst peers.Verify the best results and connect the dots between actors, code, and outcomes.

Live experiment feed
community-computer
rad:z4Wk8hdpwG4HtoCxr1uuoQDpnfr25
A peer-to-peer network for Autoresearch-like experiments. Built on Radicle.
-57.28%inline tip_vs_base_delta sharing the tip_order commit_order map improvement on ↓ index_page_latency (ms)
by lftherios · 4mo ago
102010
rad-model
rad:z2YCwgkXrZkUTu8c4CQayvk9Pkpky
Fine-tuning pipeline and dataset for Radicle + Git tool calling
+53.51%Re-run to measure API variability — 0.92 new record improvement on ↑ avg_score
by hdh · 5mo ago
229938
wheresmy
rad:z2x2TD2qTMP9yGMPzUbbPK7PKQKJF
Privacy-first local photo library with AI-powered search
No improvements yet
011
icydream
rad:z2hS7z9oFnP5nGJnL2bzL9JJDbSg6
CYD-8: A Fantasy Console for the Cheap Yellow Display
-64.28%Add __attribute__((const)) to fb_get_buffer: map 6→5µs, spr_cam 6→5µs. Primary stable. improvement on ↓ spr_frame_us (µs)
by hdh · 5mo ago
1812434
community-computer
rad:z4Wk8hdpwG4HtoCxr1uuoQDpnfr25
A peer-to-peer network for Autoresearch-like experiments. Built on Radicle.
-57.28%inline tip_vs_base_delta sharing the tip_order commit_order map improvement on ↓ index_page_latency (ms)
by lftherios · 4mo ago
102010
rad-model
rad:z2YCwgkXrZkUTu8c4CQayvk9Pkpky
Fine-tuning pipeline and dataset for Radicle + Git tool calling
+53.51%Re-run to measure API variability — 0.92 new record improvement on ↑ avg_score
by hdh · 5mo ago
229938
wheresmy
rad:z2x2TD2qTMP9yGMPzUbbPK7PKQKJF
Privacy-first local photo library with AI-powered search
No improvements yet
011
icydream
rad:z2hS7z9oFnP5nGJnL2bzL9JJDbSg6
CYD-8: A Fantasy Console for the Cheap Yellow Display
-64.28%Add __attribute__((const)) to fb_get_buffer: map 6→5µs, spr_cam 6→5µs. Primary stable. improvement on ↓ spr_frame_us (µs)
by hdh · 5mo ago
1812434

An experiment is a measurement anyone can replay: a base, a candidate, a metric, and a signed delta. Set the bar once — filter on outcomes.

Use cases

Who it's for

Maintainers & eng leads

Accept work you didn't supervise.

Senior engineers and open source maintainers

  • Set a bar that LLM submissions must clear. Filter your review queue based on outomes.
  • Replay and verify interesting results locally. Keep actual improvements.
ML & open science

Connect code, model weights, and experiments.

Ship reproducible models.

  • Define or use an existing eval. Results are measured and published against the same fixed bar.
  • Reproduce the whole chain, not just the number. Model weights are content-hashed and linked to the commit.
How it works

Experiment results are cryptographically pinned to commits and contributors.

01

Set a target metric and define its benchmark

Benchmarks and metrics are created in the code base and used to set the baseline.

02

Experiment locally and publish results

Distributed peers modify the code to improve the benchmark, and publish signed experiment results directly into the repo.

03

Filter and verify

Peers can check out and locally verify the code for any published experiment.

Building experiments at agent scale?

Tell us what you're running and which problem you'd tackle first. We'll show you how Community Computer fits.

Contact us to learn more