The Harness Engineering & Model Wrangling Hackathon
Sep 26, 2026 · New York, NY
Ripple designs circuit boards from plain English, grades them against hidden checks, and evolves its own harness from the failures, going from 0/4 to 3/4 on…
+1Offload learns the work you repeat, so next time you can hand it over. Its engine, REM (Replay · Evolve · Merge), is an agent harness with a day/night cycle…
Recursive Computer Use is a local-first desktop automation harness that enables AI agents to execute tasks on your computer, verify outcomes locally, and…
Da Vinci is a recursive improvement CAD harness designed to support physical engineering tasks through specialized AI agents, programmatic CAD and automated…
CyclopsDiary gives factory workers wearing Meta AI glasses a shared memory of the floor. Every worker's glasses feed one MongoDB memory that keeps what was…
A harness that teaches itself what context to show Jev, TypeSafe's fast decision model, proves the result on a storm it never saw, and can't cheat because…
Shibboleth is a harness trust layer for open-weight AI models. Anyone can download a model, quietly strip out its safety ("abliterate" it), and re-upload it…
Problem statement: Two, Long Horizon Engineering. Also covers One: the agent evolves its own rules. Bookie is a long-horizon agent system that makes money on…
Adaptive Pentesting Agent, is an autonomous AI agent designed to resolve the token overload and context bloat found in existing open-source penetration tools…
SubmissionAI is a long-horizon review harness for evolving drug-development document packages. It stores immutable package versions, evidence passages…
**Self-Evolving Chip-Design Harness**: an AI harness that fixes its own agents, proven where results can't be faked. Getting a chip to hit its clock speed is…
Results snapshot (read-only): https://hindsight-xi-two.vercel.app Hindsight turns real coding-agent history into a long-horizon memory benchmark on MongoDB…
Chatbots are set up once by hand and improving them takes weeks of reading transcripts. Our MongoDB Atlas support assistant improves itself. Jev makes the…
Track: Recursive Harnessing. SwipeHome is a self-improving real-estate recommendation harness built on MongoDB Atlas. Before showing a home, it predicts…
1. The problem Data center permitting in the US takes one to three years, and much of that time is rework. A site plan goes to the county, comes back with…
Simulacra is a real-time, multiplayer economic-policy simulator: a living pixel-art city where people and AI play out what a policy actually does to…
Jev-gating memory is self-improving memory for Strands agents, backed by MongoDB Atlas. A type-safe classifier called Jev decides what is worth remembering…
Chronicle is self-healing project memory. It finds conflicts in team updates. People can correct its decisions. Chronicle saves these corrections as scoped…
An AI stylist that figures out how fashion items work together — across users’ closets and retail products. It saves users time deciding what to wear and how…
Converge is collaborative coding workspace through a custom Mac app, web app, and Codex plugin that lets teams guide a shared coding agent. It tracks each…
LEVY — The Tariff Intelligence Desk Tariff wars move in hours. LEVY forecasts them like a trading desk with receipts. Every probability has a belief timeline…
Attune is an AI-powered experimentation lab for improving website flows. A business owner sets an optimization goal and captures a user journey; agents plan…
Reflexes: agents that grow reflexes. The problem: AI agents send every small decision (classify, prioritize, route, check policy) to a slow, expensive LLM…
MongoHive turns what one agent learns into tested tools every teammate's agent can call by name. When your Claude or Codex agent finishes an investigation, a…
Ada is a 3D quadruped in MuJoCo whose walking gait is written by a cheap model, gpt-6-luna. A supervisor model rewrites the harness around it (its rules…
A self-evolving health-insurance operations copilot that learns how each reviewer makes mistakes, adapts its tools, context, and coaching to that person, and…
+1Moss: long-running AI teammates whose runs and memory live in one MongoDB Atlas cluster, and that forget on evidence. You message a coordinator (HQ). It…
+1In agent-run analytics and engineering workflows, outputs must cite their evidence; when they fail, teams tune prompts and checks by hand without a fixed…
Devrandom is infrastructure for AI agents that improve how they work while remaining accountable and within user-defined permissions. A user defines a task…
Antibody is an immune system for AI agents. When an attack gets through, it writes a fix, proves the fix blocks the attack without breaking legitimate work…
+1Blanko is a better N8N that allows you to build share and manage agentic architecture
Recall Gate is a cost-rationing safety monitor for cyber-capable agents: instead of running an expensive LLM monitor on everything an agent does, it reads the…
Claim Cipher Insurers use algorithms to deny claims. We built a self-improving agent that learns each insurer's hidden behavior, catches wrongful denials, and…
long context agent
Second Shift is a long-horizon NeuroAI harness that keeps an agent pursuing one research goal across context resets, crashes, and changing constraints…
Self-Taught Trainer is a self-improving harness for an LLM that plays Pokémon Showdown (Gen 1 OU). The player, GPT-5.4 mini, is never retrained. Instead, a…
AI agents that send emails, read files, and browse the web can be hijacked by prompt injection, where an attacker hides instructions in content the agent…
+1LILI An adaptive discovery assistant for finding interesting places, local communities, and professional opportunities. Built by Ed Amaya with Python, Flask…
Bisect is a self-improving agent harness that never debugs the same bug twice. Agents that learn rules from their mistakes also learn bad habits. Bisect runs…
Proxy Harness is an evaluation harness for applying an institutional investor's proxy-voting policy to SEC proxy statements with LLMs, built for Pennant (YC…
Red vs Blue: an agent that teaches itself to attack A self-improving agent harness, shown as a game. On a simulated office network, an AI attacker plays…
Kiara is a self-improving legal-context harness on MongoDB Atlas. A California signup at DemoCo, a fictional New York startup, triggers a scoped applicability…
Here's a description you can paste straight into the form: Downshift is a self-improving harness for data questions on MongoDB Atlas. The first time a new…
An assistant gives Tony a target price, a private walk-away limit, required terms and a deadline. From there Tony runs the negotiation. It writes firm…
A rover 20 light-minutes from Earth can't wait for an engineer when its agent fails, so our harness repairs itself. An Explorer agent (Claude Haiku 4.5)…
Context Plane — Mission Control for AI Teams Context Plane connects the AI agents working across a company and gives them shared, durable context. Each user…
+1A Data layer for agents with memory context options. Reduces complexity for developers when working with vector DBs and topic targeted agents. Leverages…
A coding-agent harness that detects repeated failed fixes and changes the agent's context using evidence stored in MongoDB Atlas.
Record agentic traces, store them on MongoDB and replay them to help debug and improve the harness automatically.
Seal (Self Healing Harness) is an AI agent harness built on top of a simple Database Query Resolution Agent that learns from its own limitations. When a…
HarnessForge: Evolutionary CI/CD for operational AI agents HarnessForge continuously hardens an AI agent by turning its failures into measurable harness…
+13PT (Three Point Harness) is a self-improving harness for general contractors, built for Problem Statement One, Recursive Harnessing. Supers, PMs and project…
Relay is an agent for the weeks after a storm, when volunteers rotate out and the people who were reported once are forgotten. It listens across 311 calls…
Two companies, two private playbooks, one signed joint procedure. Neither side reads the other's. Interlock merges them on MongoDB Atlas: Vector Search aligns…
+1Try now: AiLobang.com AI Lobang is a voice-first agent you reach with one phone call. Connect your accounts- Gmail, Drive, Telegram, your bank, and it…
Long-running agents degrade because their context is a pile and they drift because the thing doing the work is also the thing remembering why. goalkeeper…
A voice first agent with the ability to generate its own capabilities and interface.
Harness Architect: an autonomous infrastructure engineer for GPU inference Owning GPUs shouldn't require becoming an inference expert. Describe your hardware…
+1Zorkinator plays Zork I with a harness that turns every death into a lesson instead of forgetting it. It builds its own map and memory as it plays, and a rule…
+1Aegis is an evidence-governed repair harness for broken API integrations. It uses Strands agents, isolated candidate workspaces, MongoDB Atlas durable memory…
+1Incident Memory prevents engineering teams from repeatedly solving the same production problems by storing past incidents, failed approaches, and human…
We built a complete orchestration and memory system for teams of coding agents. At the foundation is our custom Orchestrator Harness. Every campaign begins…
+1Journey Harness is a long-horizon AI engineering system designed to keep agents working reliably toward complex software goals across multiple steps…
An agent working for weeks should not forget what it learned yesterday. Long Horizon Memory gives agents persistent, connected memory so they can recognise…
+1Cortex is a memory for your computer that remembers what you use, lets unused detail fade, and helps an AI assistant keep up with your life. It turns…
Scar Tissue: Every agent failure becomes a tested upgrade to its own harness. An LLM agent takes a small grocery's orders in MongoDB Atlas. When a run breaks…
Pregame is a call-prep assistant for a bank advisor, with a harness that keeps its own memory honest. Advisors stop trusting AI assistants within months…
Xchange gives an engineering team's AI coding agents one shared memory. When an agent learns something useful on the job, it shares it on its own, and every…
+1Gatekeeper: a procurement agent whose guardrails evolve themselves (Statement 1: Recursive Harnessing) AI agents are starting to approve vendors, and they…
Robologue: a dialogue for robots. Procedural memory for long-horizon tasks. Long-horizon robots fail for a simple reason: they forget what went wrong five…
Visai is a self-improving agent that makes a model faster on the machine it actually runs on, then remembers what worked so the next run searches less. You…
ToolSmith is a procedural memory layer for AI agents that turns repeated work into tested, reusable tools. It watches how people and agents work, both on…
Can we stop AI agents from going rogue before they take over the world? Immune Harness is a security layer for AI agents. It checks every tool call before it…
Goodhart's law says that when a measure becomes a target, it stops being a good measure. Language models obey this pretty well; as ImpossibleBench documents…
Omega is an adaptive context and execution layer for high-stakes operations. It ingests fragmented information from sources such as email, Slack, CRM, project…
+1Large language models are a new frontier in how we interact with and generate digital content, but they still face significant limitations when working with…
Rulebook — the harness that writes its own rulebook, from the one you already wrote by hand. Every team using a coding agent has been building a recursive…
This is a cybersecurity agent that learns from past cybersecurity attacks. This means it is better prepared in the future for future attacks that are similar…
Darwin. Self-improving research system for 5S rRNA biology. This project builds a self-improving research system to investigate how changes in 5S ribosomal…
Raw Material Intelligence for Automotive Purchasing
Tomok turns heavy civil construction project history into planning knowledge.
# AgentJIT: a JIT compiler for LLM agents ## The problem Production agents pay full LLM cost and latency for every task, even the thousandth one that follows…
We built a harness for track 2 targetting memory management. It included an observer agent for managing data streams.
+1Safe Harbor: real genomic evidence investigations with durable memory and evaluated harness adaptation From docs/safe-harbor/SPECIFICATION.md: What it is: a…
PortPilot helps migrate / port a software service from one legacy tech stack to other. Legacy services are mostly radioactive and no one wants to own them and…
Flake is an AI agent that plans group events like dinners and trips. It keeps the organizer from getting stuck with the bill when friends bail or pay late, by…
Rekurse helps coding agents avoid repeating the same failed approach. When an agent gets stuck, Rekurse records the session and creates checkpoints before…
+1Frontier models can work for hours, but one hallucination can snowball into bad plans, tool calls, and wasted spend. Existing validation often catches this…
Wrasse: the harness that keeps your coding agent on the plan The problem. AI coding agents are powerful but have no discipline. You give one a goal and a…
Persistent reconciliation harness built with Google ADK, OpenAI, MongoDB Atlas, FastAPI, and React.
ClaimMemory Public adjusters and disaster-claim policyholders fight insurers over months, across hundreds of documents, with no reliable memory of what was…
Many-Lives is a persistent-world agent harness, demonstrated by a turn-based text dungeon. The harness is the product; the game is the environment. It solves…
A console for project managers to query continuous OS-level and screen-capture data over long horizons using a custom agent served from openrouter using…
KYAgent is a KYB + KYA trust infrastructure layer for AI-era commerce.
Casefile helps review large construction claim files. It organizes the record into a chronology, party map, exhibit list, and coverage report, with a source…
A flood agent whose answers are fenced in by real, versioned evidence in MongoDB Atlas. Today it refuses unsupported claims. Next, it remembers every answer…
Shadow CFO ( https://shadow-cfo.vercel.app ) is a self-improving AI financial agent whose rules live as evolving documents in MongoDB Atlas, letting it…
Continuum is a memory, policy and explanation layer for AI agents on MongoDB Atlas: it recalls what worked, follows a human-approved policy, and can explain…
