Hey HN!

Wanted to show our open source agent harness called Gambit.

If you’re not familiar, agent harnesses are sort of like an operating system for an agent... they handle tool calling, planning, context window management, and don’t require as much developer orchestration.

Normally you might see an agent orchestration framework pipeline like:

compute -> compute -> compute -> LLM -> compute -> compute -> LLM

we invert this so with an agent harness, it’s more like:

LLM -> LLM -> LLM -> compute -> LLM -> LLM -> compute -> LLM

Essentially you describe each agent in either a self contained markdown file, or as a typescript program. Your root agent can bring in other agents as needed, and we create a typesafe way for you to define the interfaces between those agents. We call these decks.

Agents can call agents, and each agent can be designed with whatever model params make sense for your task.

Additionally, each step of the chain gets automatic evals, we call graders. A grader is another deck type… but it’s designed to evaluate and score conversations (or individual conversation turns).

We also have test agents you can define on a deck-by-deck basis, that are designed to mimic scenarios your agent would face and generate synthetic data for either humans or graders to grade.

Prior to Gambit, we had built an LLM based video editor, and we weren’t happy with the results, which is what brought us down this path of improving inference time LLM quality.

We know it’s missing some obvious parts, but we wanted to get this out there to see how it could help people or start conversations. We’re really happy with how it’s working with some of our early design partners, and we think it’s a way to implement a lot of interesting applications:

- Truly open source agents and assistants, where logic, code, and prompts can be easily shared with the community. - Rubric based grading to guarantee you (for instance) don’t leak PII accidentally - Spin up a usable bot in minutes and have Codex or Claude Code use our command line runner / graders to build a first version that is pretty good w/ very little human intervention.

We’ll be around if ya’ll have any questions or thoughts. Thanks for checking us out!

Walkthrough video: https://youtu.be/J_hQ2L_yy60

Gov. Tim Walz encourages Minnesotans to film ICE agents for future prosecutions

Build Cursor Tab in less than 300 lines of Lua

The Changing Face of Koreans

Show HN: Cron for Claude Code – quickly schedule repeating CC jobs

Student arrested for eating AI art in UAF gallery protest

Uncensored General Intelligence (UGI) Leaderboard

Ask HN: How do you realistically prepare for retirement while working in tech?

Nvidia allegedly slashes GPU supply 20% – no new GeForce gaming GPU until 2027

Show HN: Lwt-to-Eio – Automated Migration from Lwt to OCaml 5 Direct Style

Will the 'Scandinavian sleep method' help me sleep?

Everything Becomes an Agent

Show HN: Gambit, an open-source agent harness for building reliable AI agents

My Gripes with Prolog

WinBoat: Drive by Client RCE and Sandbox Escape

Amazon Is Buying America's First New Copper Output in More Than a Decade

I Mapped Over 300,000 Brands

YBOOK: First Million in Crypto

Lex-Gql: GraphQL for AT Protocol

Bluetooth – Origin of the Name

List of Individual Trees

"You Had One Job": Why Twenty Years of DevOps Has Failed to Do It

HTTP RateLimit Headers

A superforecaster shares what bottom-feeders can teach about consuming media

Show HN: Camera calibration directly in browser for SLAM

All 23-Bit Still Lifes Are Glider Constructible

The Magic of the Better Software Conference

The Honey Scam Is Worse Than I Thought

Invading Greenland? Mistake of a Lifetime [video]

A Game to Align People and Priorities

Timeplus now natively support splunk S2S protocol

Show HN: Jtpck sends your Claude Code, Codex, and Gemini CLI back to you