JamJet Cloud
Production-safe AI agents. Drop-in governance, telemetry, and metered cost recovery for any AI app.
JamJet Cloud
JamJet Cloud is the hosted governance layer for AI agents. Drop in two lines of SDK and you get policy enforcement, budget tracking, approval flows, telemetry, and an audit trail, for any AI app on any framework. The dashboard organizes everything into three chapters: Observe what your agents do, Enforce what they can spend and call, and Prove that they ran safely.
Get started
Quickstart
Sign up, install @jamjet/cloud, add policy and budget. About 5 minutes.
Concepts
Span, Agent, Policy, Budget, Approval. The five primitives.
Cloud vs Open Source
Decide which JamJet you need. They work together.
Observe
The Observe chapter is where you watch what your fleet actually does. Every wrapped LLM call becomes a span with tokens, latency, and cost in USD, attributed to the agent and user that produced it. Costs and the network graph are good places to start: you can see where spend concentrates before you decide what to enforce.
Runs and traces
Every LLM call captured as a span. Model, payloads, tokens, latency, cost.
Network graph
Multi-agent topology rendered from spans. See how work flows across agents.
Agents
Named identities tagged on spans. Slice telemetry and cost per agent.
Costs
Spend broken down by model, agent, and project. A natural entry point.
Enforce
The Enforce chapter is where governance turns into action. Policies control which tools an LLM can call, budgets cap spend before billing, and approvals gate sensitive actions on a human. The cost-recovery loop lives here: Optimize turns a waste finding into an enforced policy in one click, and Savings reports the spend that enforcement actually recovered, metered from real call data rather than self-reported.
Policies
Block dangerous tools at runtime. Glob rules, pre-call filter and post-decision check.
Optimize
Waste findings you can turn into an enforced policy in one click. A good entry point.
Budgets
Per-project cost ceilings. Pre-call estimate, post-call truth.
Approvals
Human-in-the-loop gates. Block the call until someone approves or rejects.
Savings
The spend enforcement actually recovered. Metered, not self-reported.
Prove
The Prove chapter is your evidence layer. It answers the auditor's question: was this agent safe to ship, and can you show what it did. Readiness scores your project against production checks, the audit trail searches every governance decision, and datasets capture inputs and outputs for evaluation.
Readiness
Score your project against production-safety checks before you ship.
Audit
Search every blocked, approved, and pending decision across spans.
Datasets
Capture inputs and outputs for evaluation and regression checks.
Memory
Hosted shared memory across your fleet. BYOK embeddings, GDPR delete-by-user.
Integrations
Vercel AI SDK
wrapLanguageModel middleware for streamText and generateText.
All integrations
LangChain4j, Spring Boot, more.