AI engineer, building AI products,Agents, RAG, LLM apps and RLHF post training.
- Tweets 4,815
- Following 2,199
- Followers 364
- Likes 1,497
LlamaFactory made fine-tuning boring in the best way. PenguinHarness is trying to do the same for agent self-improvement: one framework for the RSI methods, plus GDPevo to compare them across 240 enterprise tasks. The part I'll be watching is how the evals stay honest while the agent rewrites itself.
👋 Hi everyone! Long time no see
We're the LlamaFactory team. After passing 70k stars on GitHub, we're now building the "LlamaFactory" of recursive self-improvement (RSI), towards a unified, efficient, stable, and flexible framework for agent self-evolution. Excited to share our team's latest work with you all.
TL;DR:
🐧 We've open-sourced PenguinHarness, a multi-agent self-evolution framework that brings every RSI algorithm under one roof, over 2,600 GitHub stars.
📖 To get a fuller picture of existing RSI work, we went through 140 papers, organized and categorized them into the awesome-rsi repo, and published a survey paper.
📊 To compare self-evolution efficiency across algorithms fairly, we built GDPevo, a self-evolution benchmark covering 6 enterprise scenarios and 240 tasks, and used it to evaluate 6 mainstream agents.
The details:
Following the same philosophy as LlamaFactory, we want to bring all RSI-related data, models, algorithms, and benchmarks into a unified open-source harness, so everyone can try and debug RSI more easily. We've already integrated 5 RSI algorithms and 7 benchmarks, and we're aiming for 100+ down the road.
Borrowing from DSH's "everything is a plugin" approach, we've packaged every algorithm and benchmark as a plugin. They're fully decoupled, so users can freely mix and match any RSI algorithm to evolve different agents. We also built a modern desktop client that installs in one click on any OS and runs locally, so you can grab the latest version from our website.
Website: penguin.ooo/
OSS Repo: github.com/Prism-Shadow/peng…
To better understand where the academic work stands today, we open-sourced the awesome-rsi repo and published a survey paper covering 140 recent works. We also mapped the trajectory of RSI against that of deep learning — many concepts line up one-to-one, only now the subject is an agent.
OSS Repo: github.com/Prism-Shadow/awes…
Paper: arxiv.org/abs/2610.14625
To better evaluate how different harnesses and methods perform at self-evolution, we built the GDPevo benchmark, spanning 6 real-world enterprise scenarios and 240 tasks. With a train/val split, it fairly measures an agent's self-evolution ability while guarding against overfitting.
OSS Repo: github.com/Prism-Shadow/GDPe…
Paper: arxiv.org/abs/2608.03764
This post isn't just about our work. We're also hoping to invite friends from academia and industry to join our open-source project. If you'd like to see your name on an open-source project and co-author papers with us, DM us and let's talk!
3
341
I built a GTM agent that remembers why a client said "stop campaign A", and knows whether that was a one-off or a rule for everything after. It is called Precedent and it is open source. This is the architecture.
2
1
15
1,304
Every message becomes a structured decision, with reason, evidence and outcome stored separately. Process is judged on what was knowable then, facts are bitemporal, every action passes a risk-tiered approval gateway, and forgetting a client returns a zero-residue certificate.
1
4
631
Code, architecture write-up and a Streamlit UI, MIT licensed. Tell me what you would break. github.com/cero753/precedent
1
48
I'm looking for full-time AI engineer or forward deployed roles, so DMs are open. Tell me what you'd break
1
28
And here is the repo with the code, the architecture write-up and a Streamlit UI, MIT licensed: github.com/cero753/precedent
12
Friends at Polygres launched WonderSearch today go try it out
Today we’re launching WonderSearch (YC F26)
Search millions of unstructured documents without pre-embedding the corpus. With comparable/better retrieval recall than embedding-based search.
WonderSearch searches at query time:
> No corpus-wide embedding step.
> No vector index to maintain.
> More compute only when needed.
> Pay per search intead of embeddings you may never use.
Go from raw data to useful answers in seconds.
Link + FAQ below (+ free launch credits!) ↓
1
5
1,269
NVIDIA open sourced OpenShell today. It is a runtime boundary for agents that traces every action and enforces policy, and the new release adds a policy prover that checks whether a set of permissions can be combined into something the operator never intended. That prover is the interesting part. Most agent sandboxes gate single calls. Real breakouts come from chaining allowed calls, read a token in one tool, spend it in another. A harness primitive that reasons about composition instead of individual tool calls is what has been missing. Sentry, the watchdog half, runs out of band on a DPU and quarantines an agent in milliseconds. Not open source, but the APIs are. Going to run my eval traces through OpenShell this week and see what the prover flags.
3
1
8
712
Most agent failures I debug are not model failures. The model picked a reasonable action given what it saw. The harness fed it a stale summary, swallowed a tool error, or handed it a permission it never should have had.Fix the loop before you swap the model.
2
1
10
800
Union Alpha is live on OpenCode and it is free.Stealth frontier-level model built for agentic coding: 262K context, 131K output, images, tool calling, zero data retention.opencode, then /connect, then /models, pick Union Alpha Free. Try it while the free window lasts.
3
862
Most real APIs don't ship an OpenAPI spec. Banks, real estate, internal company tools. Just HTML docs a human is supposed to read.
w2mcp reads those docs the way a developer would and generates a working MCP server from them, no spec required.
5
1
7
921
For teams that don't want to run their own gateway, there's a paid hosted tier. Credentials encrypted at rest, injected per request, servers kept in sync when the underlying API changes. Free tier covers local use with no cost.
1
31