Glossary

Agent

A named identity that produces spans. In JamJet Cloud, agents map to nodes in the dashboard's network graph. Created via agent('researcher') (TypeScript) or jamjet.cloud.agent('researcher') (Python).

Approval

A human-in-the-loop gate. When requireApproval('action_name') is called, the SDK polls JamJet Cloud's approvals API until a human approves or rejects in the dashboard.

A2A

Agent-to-Agent protocol. JamJet has native A2A support — agents can discover and call each other across processes and networks. See A2A protocol.

Budget

A cost ceiling enforced by JamJet Cloud. budget(50) sets a $50/project ceiling; calls that would exceed it throw JamjetBudgetExceeded.

Checkpoint

A serialized snapshot of a workflow's state. JamJet's open-source runtime checkpoints after every node completes, so a process restart can resume from the last checkpoint without re-running prior work.

Engram

JamJet's temporal-knowledge-graph memory layer. See Engram.

MCP

Model Context Protocol. An Anthropic-introduced protocol for connecting AI agents to external tools and data sources. JamJet has native MCP support. See MCP protocol.

Policy

A rule that controls which tools an LLM can call. policy('block', 'wire_*') blocks every tool whose name starts with wire_. See Policies.

Span

A unit of observed work — typically one LLM call or one tool invocation. Spans carry agent identity, user context, cost, tokens, latency. The basic unit of JamJet Cloud's telemetry.

Trace

A correlated set of spans. JamJet supports W3C trace IDs so spans across services chain into a single trace.

Workflow

The open-source runtime's primary unit of authoring. A @jamjet.workflow (Python) or @Workflow (Java) defines a graph of nodes with checkpointed transitions.

YAML Workflows

A declarative authoring surface. Defines workflows in YAML instead of code; useful for non-developer authors and for workflows that need to be edited at runtime. See YAML Workflows.