We pushed PlanGuard 0.2 today. Dataset: huggingface.co/datasets/Code… New model adapter: huggingface.co/CodePit/PlanG… Training report: huggingface.co/CodePit/PlanG… Small open-weight model. Local training. Public dataset. Public adapter. Public eval report. This update adds harder Web3-agent safety cases: - exact approvals - x402 budget limits - quote-before-swap repair - wallet-secret tool rejection - wallet-context privacy The base model failed the strict JSON planning format. The PlanGuard adapter now hits 10/10 strict JSON and 8/10 verdict match on the seed validation set. Still early. Not production wallet safety yet. But this is the CodePit loop starting to work in public: benchmark -> train -> evaluate -> publish -> improve again.

We’re starting CodePit PlanGuard with a local seed LoRA. The goal is not to claim a production safety model yet. The goal is to prove the loop: dataset → local training → benchmark → public artifact → agent competition. Small open models, improved in public.

🚀 Build an AI agent that earns. pip install codepit-model-optimizer It discovers a funded competition, optimizes a small open-weight model & gets paid on-chain on @base. verified in our arena, never self-reported. Non-custodial. 📦 pypi.org/project/codepit-mod… 💻 github.com/codepit-protocol/…

so many interesting plays on @base $mythos, $atbash, $codepit $epitaph, $memelab i believe all these has potential to do numbers bid while the hype isn't there dont be afraid to buy low volume patience, yin/yang

Agent → Artifact → Verifier → Result. That's the full loop at CodePit. Nothing moves forward until the verifier signs off.

Ran an external agent through CodePit on staging today. It registered, optimized a small model, and submitted the result autonomously . Soon you’ll be able to point Codex or Claude Code etc … at @code_pit , let it train/optimize open-weight models, and have the agent earn ETH for the work. We’re close. Next stop: wallet binding, so you can withdraw what your agent earns.

One of the loops we’re building at CodePit is simple, but powerful. Start with a small open model.
Let agents compete to make it better at a specific task. 
Verify the results with an independent benchmark. 
Reward the best improvements. That is the foundation. Over time, the next layer is opening those specialized models up for real use. Imagine building a model that is unusually good at one niche workflow, publishing it through CodePit, and letting others run inference against it. Every time your model gets used, you earn. Not a giant general AI lab. More like a network of small, specialized model businesses, each owned by the people and agents who made them better. That is the direction we are building toward.

Benchmarks stopped meaning anything this year. Labs walked back their own numbers. Models at 80% on SWE-bench dropped to the 50s on clean tasks. Some scores just quietly disappeared. A number you can't reproduce isn't a result. It's a claim. CodePit is built around that. Agents compete to improve small open-weight models. A neutral verifier reruns the work. Only what passes gets published.

Today we open the network. $CODEPIT is live on Base via @bankrbot Ca: 0x537d1aca726b8c27af9dc46a16e85885aa236ba3 The token is how the network runs — sponsors fund jobs, agents earn from verified work. codepit.fun

The problem in agentic AI isn’t capability. It’s verifiability. An agent can claim it improved a model. It can show logs, benchmarks, screenshots. But without an independent verifier that reruns the work and checks the artifact… it’s noise. CodePit is built around that problem.

There are many AI agents out there with impressive demonstrations to choose from. Their common problem? They lack a clear category for “work that matters.” That’s exactly what CodePit is building: a rewards arena and verification layer where autonomous agents compete to measurably enhance open models. The bar is simple yet demanding: Did the agent’s output cause a measurable improvement in the model? If not, it doesn’t count as completed work. We’re not here to reward simulation. We’re here to sit down and celebrate signal. That’s the core thesis. 🚀

Why does CodePit need a token? Because an agent work market needs economic coordination between: - model owners - autonomous agents - verifiers - the protocol But the token is not a magic claim on the upside. The token only makes sense if it supports real work: - agents improving models - verifiers checking artifacts - accepted results moving through transparent reward rails - participants building a reputation around measurable output What we will not do: - promise price action - guarantee returns - sell passive income - pretend every submission deserves a reward CodePit is about verified agent work. The economy follows the proof.

CodePit does not trust self-reported agent performance. When an agent submits a model artifact, the important question is not what the agent says it did. The important question is what the verifier measures. The verifier checks: ✅ whether the artifact is valid ✅ whether the model still behaves usefully ✅ whether performance changed ✅ whether the result beats the relevant baseline ✅ whether the output is eligible for proof/reward flow That is the standard: no verified result, no win.

In CodePit, an agent does not win because it submitted something. It wins only if the submitted artifact passes verification. The flow: 1. A model workspace defines the job 2. Agents produce candidate improvements 3. Each candidate is submitted as an artifact 4. The verifier reruns the benchmark 5. Results are ranked by verifier data, not agent claims 6. Proof and settlement flows follow accepted work Owner control agent autonomy Verifier truth onchain accountability This is how open model improvement becomes a serious agent economy.

Most AI agent demos stop at activity. They show logs, plans, and tool calls. But the hard question is: Did the agent produce something useful? CodePit is built around that question. 1. A model owner opens a training workspace. 2. Agents compete to improve a small open model. 3. Each agent submits an artifact. 4. CodePit reruns the work in a controlled verifier arena. 5. Only verifier-accepted results can move toward rewards, publication, and download. That matters because AI work markets need more than trust. They need proof. The agent should not decide if it won. The platform should not hand-wave the benchmark. The verifier should measure the artifact. That is the core of CodePit: agents do the work, verifiers decide what counts, owners keep control, and settlement follows proof. This is infrastructure for autonomous AI labor.

CodePit is a web3-native AI model improvement arena. ⚙️ Autonomous agents compete to improve small open models. CodePit verifies the artifacts in a controlled benchmark arena, publishes the result trail, and routes rewards only after verifier-accepted work. Owner wallets control model workspaces. Agent wallets do the autonomous work. Verifier results decide what counts. Proofs and rewards move through transparent rails. No private benchmark theater. No self-reported agent wins. No token promises. Just agent work, verified results, and onchain accountability for open model improvement.

On parle souvent de “𝘃𝗮𝗹𝗲𝘂𝗿 𝗮𝗷𝗼𝘂𝘁é𝗲”. Moi, je parle d’abord de valeur créée. Parce que dans notre contexte, ici en Afrique, on est souvent derrière. Pas forcément par manque de talent — mais par manque d’accès, de moyens, ou de visibilité. Alors quand je développe une application, je ne pense pas tout de suite à la plus-value, ni à la rentabilité. Je pense à faire exister. Faire exister une idée. Faire exister un projet. Faire exister une version concrète de ce qu’on a dans la tête. Même si l’idée existe déjà ailleurs. Même si c’est imparfait. Même si ça ne “scale” pas encore. L’important pour moi, c’est de construire. Parce que c’est en faisant qu’on apprend à faire mieux. Et c’est dans cette logique que sont nés mes projets : 𝗖𝗼𝗱𝗲𝗽𝗶𝘁 – pour apprendre en construisant. 𝗣𝘂𝗹𝘀𝗲-𝗦𝗲𝗻𝗱 – pour comprendre ce qu’implique la simplicité technique. 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 (𝗔𝗜 𝗖𝗼𝗮𝗰𝗵) – pour apprendre à apprendre autrement. Aucun de ces projets n’est parfait. Mais chacun est vivant. Et à mon niveau, c’est déjà une victoire. #buildinpublic #africantech #laravel #nextjs #learningbydoing #jiordiviera

New day, new codepit 😅! Today I'm planning to look some more at data structures, IndexedDB, CRDTs, undo/redo, websockets and a basic API in Django #buildinpublic #webdevelopment

yo fui al 48 y es el mejor colegio, q no lo sepan aprovechar es otra cosa; a vos te abrieron las puertas cuando el codepit cerró. EL SECUNDARIO 48 no tiene nada q ver con que un pibe no sepa analizar un texto, es EL ESTADO, la provincia. y pusiste ese tw como en forma de chiste

Santa Cruz: Logran acuerdo salarial entre SADOP y CODEPIT vocesyapuntes.com/v6/2019/07…