Coordinate, review and safeguard long-running AI agent work
My existing harness runs missions for hours or days, with a lead agent delegating work to sandboxed agents, independent review, and a gateway for secrets, leak checks, and logging. At ROS Humanoid Hack 2026, an agent using this harness worked continuously for 292 hours and delegated 1,421 tasks. In a delivery-routing experiment, an autonomous mission produced routes 5.65% shorter than an OR-Tools baseline.
I am seeking $5,000–$20,000 to develop and evaluate an open tool based on this work. An indicative budget at the $20,000 goal is $10,000 for development, $5,000 for compute and model API usage, $3,000 for evaluation and security testing, and $2,000 for documentation and release. At lower funding levels, the scope would be reduced accordingly.
Success would mean that others can use the tool to run and independently review autonomous missions. The project could fail if reliable coordination, security, or reproducibility proves harder than expected. The budget is an estimate, and details of the release plan remain to be determined.
There are no bids on this project.
Comments
No comments yet. Sign in to create one!