Coding agents & CLI

• herdr terminal agent multiplexer, tmux-like panes per agent, socket API for agents to spawn/read/wait on own panes, sessions survive detach/ssh.
• jcode new coding agent harness pitched on multi-session workflows + deep customizability, worth watching vs Claude Code/OpenCode.
• claude-thermos pings Claude sessions to stop prompt-cache expiry during long subagent waits; community pushback: cache TTL is actually 1hr not 5min, so ping cost may exceed savings, sanity-check before adopting.
• OmniRoute free-tier LLM gateway aggregating 43 provider pools/460+ models behind one endpoint for Claude Code/Codex/Cursor, claims 15-95% token savings via stacked compression.

Agentic orchestration & fleets

• Orca ADE running Claude Code/Codex/OpenCode/Pi in parallel worktrees with unified search + usage tracking, desktop+mobile.
• ego-lite browser built for human+agent parallel use, agents get own "Spaces" so logins/tabs don't collide, fewer tokens than driving a separate automation browser.
• Copilot cloud agent for Linear GA, assign Linear issues directly to GitHub's async background agent.
• Modal on Agent Experience infra Modal CTO argues cloud infra needs rework for agent workloads not human devs, context after Modal's $355M Series C.
• Screenpipe (YC S26) local-only screen/audio recorder building a searchable memory agents can query to auto-generate SOPs; HN wants strictly local, no cloud component.
• OneCLI OSS credential gateway, agents call APIs without ever seeing keys; community notes this is same pattern as Infisical/Varlock, not a real security advance.

LLM mechanics & models

• GLM-5.2 step change for open agents Nathan Lambert flags GLM-5.2 crossing a capability threshold for open-weight agentic use.
• Claude Fable 5 safety changes Anthropic swapped silent research-query manipulation for a classifier-based approach like other safety domains.
• Echo routes each query across a pool of open models (GLM-5.2, Kimi K2.7, etc) picking best-per-problem, claims Fable-level results at 1/3 cost; HN calls it vaporware, no benchmarks or specs shown.

Code review & dev tooling

• alibaba/open-code-review OSS release of Alibaba's internal AI code review CLI, hybrid deterministic-pipeline + LLM agent, line-level comments, fine-tuned rulesets for NPE/thread-safety/XSS/SQLi, OpenAI+Anthropic compatible.

Vibe-coding & misc

• Imba 7-year-old full-stack web language compiling to JS resurfaces on HN, question of whether it's separable from its own framework.
• worldmonitor OSS real-time geopolitical/infra intelligence dashboard, 500+ feeds, AI-synthesized briefs, dual 3D/WebGL map engines, relevant as a pattern for building your own briefing pipeline.