jarviscore-framework
JarvisCore is a runtime where AI agents operate as a fleet of equal peers. Agents discover one another by capability, execute tasks from a shared ledger, and au
What is jarviscore-framework?
JarvisCore is a runtime where AI agents operate as a fleet of equal peers. Agents discover one another by capability, execute tasks from a shared ledger, and authenticate to every external service through a zero-trust broker, never with their own keys.
Overview of jarviscore-framework
What it is
JarvisCore is a runtime where AI agents operate as a fleet of equal peers. Agents discover one another by capability, execute tasks from a shared ledger, and authenticate to every external service through a zero-trust broker, never with their own keys. Unlike frameworks that rely on a single orchestrator, JarvisCore uses a mesh architecture where every agent is a peer. Peers find each other by capability using SWIM gossip, take steps from a shared ledger (Redis), and record their actions so that if one agent stops, another can claim its step and continue the work. No agent holds a credential; the Nexus broker provides credentials at the time of the call.
Key features
- Agents discover each other by capability using SWIM gossip protocol.
- Tasks are executed from a shared ledger (Redis), enabling crash recovery and work continuation.
- Zero-trust authentication: credentials are injected by the Nexus broker only at call time, never exposed to agents.
- Supports both AutoAgent (goal-oriented) and CustomAgent (plain Python) on the same mesh.
- Includes durable memory that persists across runs, keyed to agent role and scoped per tenant.
- Provides evidence receipts (
result[“evidence”]) showing each tool’s input and output. - Supports human-in-the-loop approvals for irreversible actions via
Mesh.resolve_hitl().
Use cases
- Multi-agent research teams: a scout reads official docs, an analyst compares, a writer delivers a brief with sources.
- Reading and processing entire websites: a goal-oriented agent walks pages, computes figures, and saves a report.
- Data analysis with code execution: an agent writes and runs code in a sandbox to compute figures from CSV data.
- Persistent memory across sessions: facts stored in one run are recalled in later runs without re-input.
- Crash‑resilient workflows: if a process is killed mid‑workflow, finished steps are recovered from Redis and only unfinished work is re‑run.
- Team collaboration in Slack: agents change shared files and propose changes that people accept with one tap.
Getting started
bashpip install jarviscore-frameworkjarviscore initcp .env.example .env # add AZURE_API_KEY, CLAUDE_API_KEY, GEMINI_API_KEY or LLM_ENDPOINTjarviscore check --validate-llmThen run your first agent. See the repository README for the full walkthrough.
License and pricing
License: Apache-2.0. Pricing: open source.
Based on the project's public README and repository metadata.