Platform

Four pillars, one stack

Governed agent programs for real work: durable cases, explicit lifecycles, one path to models and tools, and real pilot vs production separation (Test / Live).

At a glance

Sentient Agents

Multi-tenant runtime: cases, events, routing, tools — built for weeks-long work, not one-off chats.

Lifecycle Designer

Define triggers, stages, waits, and exits so “what happens next” is explicit and reviewable.

AI Gateway

One place for LLM and tool calls: keys, quotas, retries, and spend visibility.

Pilot vs production

Separate data, endpoints, and credentials — promotion is deliberate, not a toggle. (Environments are named Test and Live in the product.)

How it fits together

One stack: environments at the top, shared configuration in the middle, governed execution at the bottom.

Live stack walkthrough

Highlights cycle; click a step to explore

Where you work

Pilot vs production

Focus

Test

Build, draft, pilot

Live

Production traffic

What you configure

Programs over time

Sentient Agents

Registry, events, routing, tools

Lifecycle Designer

Stages, waits, exits

How execution happens

Runtime + one AI path

Agent runtime

Tenant-isolated execution

AI Gateway

Models, tools, policy, metering

Same definitions in pilot (Test) and production (Live) · promotion is explicit · all model traffic through one gateway.

1 · Sentient Agents

The platform

Built for long-running programs — not disposable sessions.

  • Durable work

    Waits and SLAs live in scheduled workflows — not only in a chat window.

  • Case as truth

    Loan, order, dispute, renewal: one record; channels are views on it.

  • Events & policy

    Real-world signals route into the right program with rules you control.

2 · Lifecycle Designer

How the program unfolds

Define when the agent acts, waits, escalates, and finishes — in a structured form teams can review before it runs.

  • Preview — See stages and triggers before promotion.
  • Thin flow, rich agents — Stages stay clear; nuance stays in the agent, not in spaghetti branches.

3 · AI Gateway

Models, tools, and spend

All LLM and tool traffic goes through one layer: credentials, retries, accounting, and limits — so usage is attributable and controllable.

  • • No ad hoc provider stacks per agent.
  • • Budgets and quotas attach to org and environment.

4 · Pilot vs production (Test / Live)

Real environments

Think pilot vs production: Test and Live differ by API target, data, and credentials — not a single database flag. Promotion is explicit.

DimensionPilot (Test)Production (Live)
IntentBuild, pilot, sandboxReal customers
DataIsolated partitionProduction data
Keys & quotasNon-prodProduction

Want the full picture?

Architecture, security, or a walkthrough — we're happy to go deeper in conversation.