JarvisCore Documentation¶
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.
New in 2.3: multiplayer
A team and its agents can now work on the same files from Slack or your own product: people ask in their own words, agents propose changes, and people accept them as themselves. Start with Build a Multiplayer Agent, or read the concept first. Everything else that changed is in the changelog.
See it in two minutes¶
Agents finding each other by capability, claiming steps from a shared ledger, and calling services through Nexus.
Quickstart¶
One LLM provider key is the only required configuration.
pip install jarviscore-framework
jarviscore init
cp .env.example .env # add your LLM key
jarviscore check # verify dependencies
The Getting Started guide takes it from here: your first agent, what comes back, and how to read it.
Explore the docs¶
Getting StartedInstall, add one model key, and run your first agent.
ConceptsAgents, the Kernel, memory, the mesh, Nexus and multiplayer.
GuidesWorkflows, browser, search, credentials, people in the loop, production.
ReferenceEvery setting, the APIs, the CLI and troubleshooting.
CompareHow JarvisCore differs from CrewAI, LangGraph and Hermes Agent.
IntegrationsPrebuilt actions for the services your agents use.
The integration catalog for your installed version is one command away: jarviscore atom list.
Run the examples¶
Each is one short file in examples/showcase, verified on a real model. Longer, multi-agent examples are in Examples.
| Example | What it shows | Needs |
|---|---|---|
| Read a whole website | An agent works a real browser through every page of a catalogue and computes the figures. | jarviscore-framework[browser] |
| Compute, never estimate | Figures from a 400-row CSV, computed by code the agent writes and runs. | A model key |
| Remember across runs | Told once; a new process, given nothing, still knows. | A model key |
| Survive kill -9 | Kill it mid-workflow; the rerun finishes without repeating done work. | REDIS_URL |
| A research team | A scout and an analyst deliver a cited brief from the live web. | A model key |
| Work as a team | A room in Slack where people ask, agents propose, and people decide. | New in 2.3 |
Community¶
Questions, showcases and live office hours are on Discord. Code and issues are on GitHub, where a star helps other developers find JarvisCore. For the overview and blog, see the developer hub; for teams, Enterprise and pricing.
