Concepts¶
Use this section to build the mental model behind JarvisCore before choosing implementation details. Read the stages in order if you are new to the framework; jump directly to a concept when diagnosing an existing system.
1. Runtime foundations¶
-
Architecture
See how the agent profiles, Mesh, durable state, integrations, credentials, and observability fit together.
-
Agents
Understand agent identity, capabilities, lifecycle, and the boundary between autonomous
AutoAgentand deterministicCustomAgentexecution.
2. Reasoning and model behavior¶
-
Planning
Learn the single-agent Plan, Execute, Evaluate loop and when replanning occurs.
-
Language Models
Understand provider configuration and the model roles used by the runtime.
-
Decision Models
Use TypeSafe Jev for bounded, typed judgments inside agent workflows.
-
Model Routing
See how capability tiers select models for reasoning, coding, browser work, and lightweight evaluation.
-
Agent Personas
Add structured role intelligence and operating context without replacing an agent's executable contract.
3. State and coordination¶
-
Memory
Follow working, episodic, long-term, and optional Athena semantic memory across turns and sessions.
-
P2P Communication
Learn how agents discover peers, exchange requests, and communicate across processes with SWIM and ZMQ.
4. Tools and trust boundaries¶
-
Contracts and Boundaries
Separate prompt instructions from execution shape, capability authority, output validation, result envelopes, and durable obligation truth.
-
System Bundles and Atoms
Understand typed integration functions, registry discovery, versioning, and how atoms differ from MCP servers.
-
Nexus
See how provider credentials are resolved at the execution boundary and kept out of agent reasoning.
Once this model is clear, move to the implementation guides.