Together AI released Together Compiler, a system in which AI agents continuously re-tune the inference forward pass as production traffic shifts. Its core is a persistent serving engine that hot-reloads agent-written forward passes without restarts, backed by a 12-layer fast screening tier and a 60-layer slow tier that confirms gains on real traffic. Together claims up to 2.6x TPS over the best open configurations on SemiAnalysis InferenceX, though some gains are marginal and all figures are self-reported; the system is closed source and already live on its serverless endpoints.
Hot Reloading
热重载
Dynamically updating code or configuration without restarting services or interrupting sessions
Curated Intelligence
Open-source coding agent OpenCode has released its stable 2.0 version, migrating its entire tech stack to Node.js/Electron. This major overhaul breaks all legacy plugins and integrations built on the 1.x API, signaling a strategic shift from lightweight CLI tools to a robust desktop application runtime.
LangChain has significantly revamped the 'Skills' mechanism in its Deep Agents framework, enabling tools to be bound directly to skill folders. This ensures tool schemas are only loaded into context when the agent reads the specific skill, optimizing context window usage and preserving prompt caches. The update also supports runtime pinning for pre-loading instructions before the first model call and allows long-running agents to hot-reload added or edited skills mid-thread without restarting the session.