Delegate work to sidecar agents that pass an approval script
infaiw is an open-source daemon-first agent harness. Delegate work to sidecar agents that must pass an acceptance script before they can report "done" — the parent decides what done means, not the model. Sessions are durable timelines you can branch from any point. Approvals wait for you even after you close the terminal. Switch between DeepSeek, OpenAI Codex, and Claude Code mid-session. MCP servers, skills, evals, and image support built in. One Go binary, under 100 MiB, Apache-2.0.
👋 Hey Product Hunt! I'm Dipankar, the maker of infaiw.
The problem: AI agents love to say "done". They hallucinate success, skip edge cases, and hand you an answer you can't trust without re-checking everything yourself. If you close the terminal, the session dies. If you want to branch a conversation, you start over.
The fix: infaiw is a daemon-first agent harness where subagents (sidecars) must pass a bash acceptance script before they can report back. The parent writes the script, the parent picks the turn budget — the model never grades its own work.
What sets it apart:
• Sidecar loops — delegated agents gated by an acceptance check, not self-assessment
• Durable sessions — close the terminal, come back later, the work continued
• Branch from anywhere — wrong turn 40 minutes in? Fork the timeline, keep both
• Approvals that wait — Allow / Deny / Deny With Reason, even hours later
• Multi-model — switch between DeepSeek, OpenAI Codex, and Claude Code mid-session
• MCP, skills, evals, images — all built in
• Under 100 MiB — 3.3× less memory than an idle Claude Code
54 days of building in public so far. I'd love to hear what you'd want from an agent harness — fire away! 🔥
About infaiw on Product Hunt
“Delegate work to sidecar agents that pass an approval script”
infaiw launched on Product Hunt on November 17th, 2026 and earned 72 upvotes and 2 comments, placing #31 on the daily leaderboard. infaiw is an open-source daemon-first agent harness. Delegate work to sidecar agents that must pass an acceptance script before they can report "done" — the parent decides what done means, not the model. Sessions are durable timelines you can branch from any point. Approvals wait for you even after you close the terminal. Switch between DeepSeek, OpenAI Codex, and Claude Code mid-session. MCP servers, skills, evals, and image support built in. One Go binary, under 100 MiB, Apache-2.0.
On the analytics side, infaiw competes within Productivity, Artificial Intelligence and GitHub — topics that collectively have 1.2M followers on Product Hunt. The dashboard above tracks how infaiw performed against the three products that launched closest to it on the same day.
Who hunted infaiw?
infaiw was hunted by Dipankar Das. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
For a complete overview of infaiw including community comment highlights and product details, visit the product overview.
👋 Hey Product Hunt! I'm Dipankar, the maker of infaiw.
The problem: AI agents love to say "done". They hallucinate success, skip edge cases, and hand you an answer you can't trust without re-checking everything yourself. If you close the terminal, the session dies. If you want to branch a conversation, you start over.
The fix: infaiw is a daemon-first agent harness where subagents (sidecars) must pass a bash acceptance script before they can report back. The parent writes the script, the parent picks the turn budget — the model never grades its own work.
What sets it apart:
• Sidecar loops — delegated agents gated by an acceptance check, not self-assessment
• Durable sessions — close the terminal, come back later, the work continued
• Branch from anywhere — wrong turn 40 minutes in? Fork the timeline, keep both
• Approvals that wait — Allow / Deny / Deny With Reason, even hours later
• Multi-model — switch between DeepSeek, OpenAI Codex, and Claude Code mid-session
• MCP, skills, evals, images — all built in
• Under 100 MiB — 3.3× less memory than an idle Claude Code
It's fully open source, Apache-2.0: https://github.com/dipankardas011/infai
54 days of building in public so far. I'd love to hear what you'd want from an agent harness — fire away! 🔥