LightRSI
A modular runtime for recursive improvement in long-running LLM agents.
Build capabilities once and run them across OpenClaw, Codex, Claude Code, and future agent hosts.
Featured plugin: TokenPilot — a cache-aware context runtime that reduces long-session token usage and cost.
vs. Vanilla · Claw-Eval continuous
vs. Vanilla · Claw-Eval continuous
vs. Vanilla · PinchBench continuous
OpenClaw · Codex · Claude Code
Plugin platform for long-running AI agents
LightRSI separates improvement capabilities from shared runtime infrastructure and host integration, so each capability can be built once and evaluated consistently.
Available plugins
Each plugin provides a distinct capability. Install what you need, skip what you don't.
| Plugin | Capability | Status | Hosts |
|---|---|---|---|
| TokenPilot | Cache-aware context management | Stable | OpenClawCodexClaude Code |
| Memory Plugin | Long-term memory | Experimental | TBD |
TokenPilot: cache-efficient context
TokenPilot is the first official LightRSI plugin. It reduces token usage and cost in long-running agent sessions through stable-prefix rewriting, context reduction, and lifecycle-aware eviction.
🔒 Stable Prefix
Rewrites developer-level context into a cache-stable form, dramatically improving cache hit rates across turns.
✂️ Context Reduction
Trims oversized tool output before it pollutes later turns, keeping the context window lean without losing signal.
🗑️ Context Eviction
Limits how much old context is carried forward as sessions grow, preventing unbounded context bloat.
📊 Reports & Visuals
Built-in inspector shows token savings, cache efficiency, and session metrics in real time.
Ready to reduce your agent costs?
Install LightRSI and TokenPilot in under 5 minutes.
Get Started →