Prismer Cloud launches long-running agent harness

This looks less like a thin wrapper around LLM calls and more like an operating layer for agents that need to survive beyond a single prompt. The pitch is credible because it targets the messy parts teams keep rebuilding by hand, but the “evolution” story will need real proof before it becomes more than branding.

• –The repo frames the problem clearly: long-running agents fail when they lose context, forget prior runs, and can’t coordinate or recover reliably.
• –The stack is unusually complete for an open-source release: context compression, working and episodic memory, retry/backoff orchestration, agent messaging, and Ed25519-based security.
• –Shipping SDKs in TypeScript, Python, Go, and Rust, plus an MCP server, lowers integration friction for existing AI coding workflows.
• –The self-hostable setup matters: teams can run it locally or in their own infra instead of depending on a black-box hosted service.
• –The real differentiator is the “evolution” layer, but that is also the hardest part to get right; durable gains will depend on whether it improves outcomes, not just stores more state.