AutomatosX – Multi-agent framework with persistent memory for developers (github.com)

AutomatosX (v5.1.0, production ready Oct 2025) is a CLI-first, provider‑agnostic multi-agent orchestration framework that gives developers a persistent, local “team” of specialized AI agents (works with Claude, Gemini, OpenAI). It turns stateless assistants into coordinated, memoryful agents: every conversation is auto-saved and automatically retrieved so agents can delegate tasks, pick up work across days, and avoid repeated context. Integration options include a Claude Code slash command (/ax:agent) for interactive dev sessions and a standalone CLI (ax run ...) for automation, CI/CD and scripting — both share the same memory and session system. Technically, AutomatosX uses SQLite FTS5 for 100% local, zero-cost full-text memory (search <1 ms, claimed ~62× faster than prior vector methods) with automatic context injection; no embedding or external API calls required. It offers natural-language delegation (multiple syntaxes, parallel execution), cycle detection, session tracking, workspace isolation, role-based delegation controls (maxDelegationDepth, role ownership introduced in v5.0.12), and an MCP server for native Claude Code integration (16 MCP tools, JSON‑RPC 2.0, ~300 ms p50 latency, <1.5 s cold start). For AI/ML teams, this means faster, private, reproducible workflows, lower cost scaling (small memory footprint), and safer cross-agent orchestration for complex engineering tasks.