What Runtime delivers
When AI tools don't share context, every human on the team becomes a router. You're the one forwarding outputs, re-briefing agents, and keeping track of where things stand.
Connect replaces that with a shared layer that all your tools — and your team — read and write from automatically. The coordination happens. You just don't have to do it.
Isolated execution
Every agent run executes in an isolated environment — no shared state, no cross-contamination between workflows or tenants. The agent runs inside Runtime's execution model, not inside your application process.
Policy enforcement before the first step
The same 25+ policy categories available in Observe — cost, safety, PII, compliance, identity, rate limits — are enforced natively in Runtime. Policies gate each step before it executes. Enforcement isn't a layer you add. It's how the environment works.
Kill switches at every level
Stop any agent, any workflow, any session — immediately. At the agent level, the workflow level, and the session level. No graceful shutdown. No waiting.
Durable workflows
Workflows checkpoint automatically at every step. A network failure doesn't lose the run. A policy trigger that requires human input pauses execution and waits — then resumes from the exact point it stopped, not from the beginning. Agents survive infrastructure restarts, model timeouts, and anything else production throws at them.
A dashboard after the fact is not governance. It's an autopsy.
Observability tools record what agents did. Waxell Observe captures that — and enforces policy during execution. Runtime goes further: it is the execution. There's no separation between running the agent and governing it.
Where LangSmith and Langfuse give you traces and dashboards, Waxell Observe adds enforcement. Runtime eliminates the distinction entirely — governance is the execution model, not a check that runs alongside it.
Two decorators. Full governance.
No instrumentation step. No wrapper classes. Define your agent with Waxell's SDK decorators — Runtime handles isolation, checkpointing, audit trails, and policy enforcement from the first run.
Every @decision is logged, policy-evaluated, and audit-trailed automatically. Every @workflow is durable and resumable. Nothing else required.
FAQ
What is Waxell Runtime?
Waxell Runtime is a governed execution environment for AI agents built with the Waxell SDK. Unlike observability tools that layer governance onto existing agents, Runtime treats governance as a first-class part of execution — every workflow step is policy-gated, every decision is auditable, and every run is isolated by default.
How is Runtime different from Waxell Observe?
Observe instruments agents you've already built — it captures what they're doing and enforces policies during execution. Runtime is the environment you build inside: governance is native to every step, with no instrumentation step and no gap between execution and enforcement.
What kinds of workflows is Runtime designed for?
High-risk, high-consequence automation: financial transactions, healthcare workflows, infrastructure operations. Runtime is the right choice when a single unaudited execution carries compliance, financial, or patient safety consequences.
How do durable workflows work in Runtime?
Runtime checkpoints workflow state at each step. If a step fails, the workflow resumes from the checkpoint — not from the beginning. Human-in-the-loop decisions pause execution until approved or rejected, then resume automatically.
Does Runtime work with LangChain or other frameworks?
Runtime uses the Waxell SDK — agents are built with Waxell's Python decorators rather than another framework's primitives. If you have existing LangChain, CrewAI, or custom agents, start with Observe. Runtime is the path for new builds or planned migrations.
What compliance profiles does Runtime include?
Runtime's policy engine includes HIPAA, SOC 2, and PCI-DSS compliance profiles out of the box. Enterprise tenants can configure data residency in US East or EU West at onboarding.
How does Runtime relate to Observe and Connect?
Three products, one governance platform. Connect governs agents you didn't build — no code changes required. Observe instruments agents you've built — two lines of Python. Runtime is what you build inside when the stakes require governance to be native to execution, not layered on top of it.