VT Code turns GPT-6 Astra into a coding agent for long-running software tasks. Its Rust-based harness manages tools, context, sandboxing, permissions, planning, and verification so Astra can work through complex multi-step changes in a controlled, reviewable way. VT Code also supports durable sessions, local models, 31 providers, and open protocols like MCP and ACP.

What real task does your product handle with GPT-6 Astra?

VT Code uses GPT-6 Astra to autonomously handle long-running software engineering tasks: exploring a codebase, planning changes, editing multiple files, running tests, diagnosing failures, and iterating until the task is complete, while VT Code manages context, sandboxing, permissions, and verification.

Hi Product Hunt, I’m Vinh, maker of VT Code, an open-source coding-agent harness written in Rust. I started VT Code as a way to understand what makes coding agents work beyond the model itself. The model reasons, while VT Code handles tools, context, sandboxing, permissions, planning, and verification. For the GPT-6 Astra Challenge, I’m testing how far Astra can go on long-running software engineering tasks when paired with a controlled, reviewable harness. VT Code supports sandboxed execution, durable sessions, human approval, open protocols like MCP and ACP, and 31 model providers. Would love to hear what you think, especially around agent architecture and long-running reliability. Thanks for checking it out! Vinh

What real task does your product handle with GPT-6 Astra?

VT Code

Open-source coding agent harness built for long-running work

Open-source coding agent harness built for long-running work