hermes-agent adds idle-triggered compaction to slim stale context before first reply
NousResearch · x · 2026-08-28
Nous Research's hermes-agent merged PR #69360 by teknium1, introducing opt-in, time-based context compaction.
- Mechanism: when a session resumes after more than compression.idlecompactafterseconds of inactivity, it compacts accumulated history up front, before the first reply — so a long-lived gateway thread resumed hours later stops re-reading its full stale context every turn (#27579).
- Implementation: an idle trigger in buildturncontext with a cheap wall-clock gap pre-check gating the token estimate; the predicate shouldidlecompact skips below the post-compression target threshold and respects the failure cooldown.
- Disabled by default (0), leaving existing behavior unchanged.
The design targets revisited sessions; Nous Research explained that compressing immediately after a response isn't done due to race conditions and because user intent helps guide compaction.
More from coding & agent
- Nanjing Univ. Releases Procedura: Agentic 3D Modeling with Procedural Control — nanjinguniv · 2026-08-28
- Agent tool calls fail silently with no traces, breaking production pipelines — Icy-Weakness8310 · 2026-08-28
- Helm MCP: Give AI assistants access to real Helm chart data, stop hallucinations — modelcontextprotocol · 2026-08-28
- SymPy Sandbox MCP: Secure symbolic math computation for LLMs via SymPy — modelcontextprotocol · 2026-08-28
- Survey on LLM Agent Evaluation: Taxonomy and Enterprise Challenges — kalyan_kpl · 2026-08-28
- Building a Sci-Fi Movie RAG Agent in ~60 Lines of TypeScript — mastra_ai · 2026-08-28