JarvisCore

JarvisCore is an open-source Python framework for building autonomous multi-agent AI systems. It provides two execution profiles: AutoAgent, which runs an internal observe-orient-decide-act loop, and CustomAgent, which exposes the execution loop for more deterministic control. Both profiles use the same underlying runtime infrastructure.

The framework is designed for agents that need to operate across extended workflows rather than produce a single response. Its memory model combines a working scratchpad, episodic ledger, compressed long-term summaries, and optional cross-session semantic memory through Athena MemOS. Agents can coordinate through a peer-to-peer mesh using SWIM gossip and ZMQ, allowing the same code to run in one process or across distributed machines.

JarvisCore also connects agents to external services through typed Python integrations called atoms. The documented catalog includes services such as Slack, GitHub, Zoom, SAP, NetSuite, Microsoft Graph, and Salesforce, while the Nexus credential layer manages OAuth2, API keys, and basic authentication outside the agent's reasoning context. Tool calls, agent turns, and language-model requests are traced automatically, with Redis PubSub, JSONL, and Prometheus outputs described for streaming, compliance, and operational dashboards.

The project includes workflow orchestration, crash recovery, and human review controls. In the financial intelligence example, three AutoAgent specialists fetch market data, analyze it, and produce a Markdown briefing in dependency order; rerunning the workflow skips completed steps. A quickstart requires installation of the Python package, initialization, an LLM provider key, and a dependency check before running an agent.