Skip to content
HUIAHUB

Agents & automation

JarvisCore

Open-source Python framework for multi-agent systems with peer-to-peer coordination, layered memory and a credential broker.

Verified Open source
Visit official site (opens in a new tab)

What JarvisCore does

JarvisCore is an open-source Python framework from Prescott Data for building systems of autonomous AI agents. The core is licensed under Apache 2.0, needs Python 3.10 or later and installs from PyPI as jarviscore-framework. One LLM provider key is the only required setting.

There are two agent types. AutoAgent runs a full observe, orient, decide, act loop and, according to the docs, can write its own integrations and repair them in a sandbox. CustomAgent exposes the execution loop so you control each step. Agents find and message each other over a peer-to-peer mesh (SWIM gossip over ZeroMQ) instead of a central orchestrator, and the same code runs in one process or across machines.

Memory has four tiers: a working scratchpad, an episodic ledger, LLM-compressed long-term summaries and optional cross-session semantic memory through a service called Athena. A credential layer called Nexus handles OAuth2, API keys and basic auth so raw secrets stay out of the agent's reasoning. Every turn, tool call and LLM request is traced, with output to Redis PubSub, JSONL files and Prometheus. A human-in-the-loop queue routes decisions that cross a confidence threshold to a review inbox. Prebuilt actions cover services such as Slack, GitHub, Zoom, SAP, NetSuite, Microsoft Graph and Salesforce.

The getting-started guide lists Anthropic Claude, Google Gemini, Azure OpenAI and local or vLLM endpoints as model options. Redis is needed only for distributed workflows, and Docker only for some local infrastructure commands. A commercial JarvisCore Enterprise edition adds access controls, model routing, managed deployment, an uptime SLA and support, priced on request.

  • Building multi-agent systems that coordinate peer to peer across processes or machines
  • Running long-lived agents that keep memory across sessions
  • Letting agents call third-party APIs without exposing raw credentials to the model
  • Adding human review to agent decisions through a review queue

Who should use it

  • Python developers building multi-agent systems meant to run unattended
  • Teams that want tracing, credential isolation and human review built into the framework

Who should look elsewhere

  • People looking for a no-code agent builder or hosted product
  • Projects on Python versions older than 3.10

Strengths

  • Apache 2.0 core with commercial use allowed
  • No central orchestrator: agents coordinate over a gossip mesh
  • Built-in tracing to Redis PubSub, JSONL and Prometheus
  • Credential broker keeps OAuth tokens and API keys out of agent reasoning

Limitations

  • Distributed workflows need Redis, and some local services need Docker
  • Model routing, access controls and the uptime SLA are part of the paid Enterprise edition
  • Small community so far: the GitHub repository had 68 stars when checked in October 2026
  • Cross-session semantic memory depends on the separate Athena service

Sources