Haven: A Minimal Coordination Layer Lets AI Agents Hand Off Work Across Runtimes
FinancialTopic6522 · reddit · 2026-09-06
A Reddit developer unveiled Haven, a temporary coordination layer for AI agents built on one idea: agents should have somewhere to go when their own runtime isn't enough.
Even an agent with tools, memory, MCP, filesystems and databases eventually needs to reach outside itself: calling another agent's capability, handing off work, requesting independent evidence, waiting on events, leaving state for agents that don't exist yet, or coordinating without ad-hoc shared channels.
Haven keeps its primitives deliberately small — LOOK → HANDOFF → GARDEN → EVIDENCE → TRAIL — with a WAKE primitive in progress so an agent can declare "I'm waiting for X," yield, and resume when the condition is met instead of polling.
The author stresses it's not another agent framework: your agent stays yours, coordination is bounded and observable, with no hidden channels or raw secrets. They're recruiting people with agents already running, asking them to point their agents at Haven and let the agents decide what to do.
More from coding & agent
- Agent Astra ports three.js demos to Godot in minutes, surprising one skeptical dev — teortaxesTex · 2026-09-06
- Dev builds surprisingly fun isometric puzzle-combat game entirely with GPT-6Astra High — IanArawjo · 2026-09-06
- Armin Ronacher: reasoning levels live in the system prompt, and switching them busts your KV cache — mitsuhiko · 2026-09-06
- mitsuhiko: switching reasoning levels busts KV caches; Fable 5.1 is the exception — mitsuhiko · 2026-09-06
- OpenAI quietly kills Codex infinite-usage bug; user burned 30-40x quota in 8 hours — returnity · 2026-09-06
- A Two-Stage Trick for GPT-6 Astra: Build Demos on Medium, Polish on Max — omarsar0 · 2026-09-06