AIOS
赞助 · Sponsor
imklom.im
AI 做图 · 一句话生成高质量图片 / AI image generation.
THE AIOS BLOG
FEATURED
赞助 · Sponsor
imklom.im
AI 做图 · 一句话生成高质量图片 / AI image generation.
Notes from the local agent layer
Deep dives, release notes, and field reports on memory, orchestration, and verification from the team building AIOS.
v5.8.2: Plans stop jumping the gun, and AIOS now speaks WorkBuddy
AIOS v5.8.2 stops plans from advancing themselves when a subagent reports success without a task id, and adds WorkBuddy as a fully-supported client — native instructions, MCP, 24/24 skills, and harness driving via the bundled codebuddy CLI.
All posts 50 articles
v5.8.1: No More Frozen Agents — aios-shell Stall Fix and LLM-Judged Requirements Clarification
AIOS v5.8.1 fixes the aios-shell MCP stall that froze opencode/codex mid-command, and replaces regex-based ambiguity detection with LLM semantic judgment — grilling happens in-execution, one decision at a time.
v5.8.0: AIOS Learns Safely — Session Memory, Evidence Gates, and Governed Self-Evolution
AIOS v5.8.0 closes the broken memo trigger chain and adds a governed self-evolution pipeline with deterministic acceptance checks, canary promotion, rollback, and version update notices.
v5.6.1: Plan-Driven Multi-Agent Dispatch — aios work Reads Your Plan
v5.6.1 makes aios work plan-driven: eligible tasks from the active structured plan become parallel work items with dependencies, owned paths, and acceptance criteria — no more implicit decomposition.
v5.6.0: Parallel Multi-Agent Coding with One Command — aios work
v5.6.0 adds aios work: one command turns any task into a planned, concurrent multi-agent dispatch — plan, implement, review, and security-check with a merge gate, instead of waiting on a single agent.
Graph Engine Locally: AIOS Composes Loop Engineering Into a Verifiable Agent Graph
How AIOS implements the Graph Engine locally — composes the Loop Engineering toolkit (verifier, exit conditions, state files) with graph nodes, edges, shared state, and failure routing into a verifiable agent graph. Compared with LangGraph, CrewAI, AutoGen.
v5.5.1: Evidence-Driven Agent Lifecycle Promotion
v5.5.1 removes the hardcoded Agent promotion bottleneck: every canonical role is smoke-tested by default and verified managed evidence promotes it into live workflow orchestration.
No posts match this filter yet.
19