Production-safe
AI agents.

Two lines. Then your agent is governed.

import jamjet.cloud as jamjet
from openai import OpenAI

jamjet.configure(api_key="jj_xxxxxxxxxxxx", project="my-agent")

# Every OpenAI / Anthropic call is now captured automatically.
OpenAI().chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "hello"}],
)

Open app.jamjet.dev/dashboard/traces — the call appears within ~5 seconds with model, token counts, cost, and duration.

What's in Cloud today

• Trace dashboard. Every LLM call captured with token counts, cost, latency, model, status. Auto-instruments OpenAI and Anthropic.
• Policy filtering. Block tools by name pattern; require human approval for sensitive actions.
• Budget caps. Hard spend limit per process; SDK raises before the call goes out.
• Approval queue. Pause execution until a human approves in the dashboard.
• Append-only audit log. Every policy decision and approval recorded.
• Cost analytics. Per-project, per-model breakdowns.
• Three-tier rate limiting. Per-IP, per-API-key, per-IP — env-tunable for self-hosted.

What's coming

• Multi-agent network graph (Q3 2026). Live force-directed view of how your agents communicate — nodes are agents and MCP servers, edges show calls, costs, error rates. Drill down to traces.
• Java cloud SDK (Q3 2026). Same drop-in for Spring AI / LangChain4j developers. Auto-instruments ChatClient, ChatModel.
• Cross-agent trace propagation. When agent A calls agent B (over A2A or MCP), traces link automatically. W3C traceparent + custom tracestate.
• Hosted Engram, bundled. Memory becomes a Cloud feature, scoped per agent, shared across an agent fleet. Open-source Engram local mode stays free.
• Centralized policy with AIP delegation chains. Visualize who-authorized-what. Compliance-grade audit.
• Replay. Re-run any trace with input recordings. Ideal for debugging incidents or testing prompt changes.
• OTel GenAI ingestion. Point your existing Phoenix / OpenLLMetry / Langfuse-instrumented apps at JamJet without an SDK migration.

What makes Cloud different

• Polyglot control plane (when Q3 ships). Python (Rust runtime) + Java (embedded runtime) — one dashboard. No competitor does this. LangSmith is Python-only and LangChain-coupled.
• Built around agent identity. Agents are first-class entities, not just spans. Agent Cards, A2A, native MCP server — folded into the dashboard, not bolted on.
• Drop-in governance, not observability. LangSmith owns observability. Cloud owns policy, audit, and approval as a service — for compliance and security teams, not just data science.
• Vertical from runtime to dashboard. JamJet runtime + JamJet Cloud are one vendor. Audit trails are end-to-end, not stitched.