Golem is the durable agent runtime. From per-user AI agents to multi-step business workflows, here's what it's built to run — and the architectural reasons why.
AI agents
Featured agent use cases
Customer support agents
A durable agent per conversation that survives crashes, redeploys, and migrations — held open for hours or days across chat, email, voice, and ticketing.
Examples
Support triage agent that resolves common tickets and escalates the rest
E-commerce concierge handling refunds, shipping status, and order changes
SRE incident copilot that triages alerts, runs playbooks, and pages humans
Per-conversation chatbot for SaaS customer success and onboarding
Coding & dev agents
Agents that read, write, test, and ship code — running for hours, executing untrusted AI output, pausing and replaying without losing plan state.
Examples
Autonomous PR agent that reviews code and proposes fixes
Prompt-to-application builder spinning up backend and frontend together
Legacy-code modernization across COBOL, ABAP, and APEX targets
CI/CD remediation agent that diagnoses failures and proposes patches
Internal data copilots
Agents inside an enterprise system of record, where multi-hour reasoning, durable RAG ingestion, and per-user permissions break naive implementations.
Examples
HR / recruiting copilot searching profiles and drafting outreach
FP&A planning copilot with NL-to-SQL across financial systems
Internal docs and knowledge agent grounded in wiki and tickets
Investment-research agent over filings, reports, and analyst notes
Per-user agent fleets
One durable instance per user, tenant, or device — millions of stateful agents that must be cheap, isolated, and individually addressable at scale.
Examples
Per-cardholder recommendation agents across millions of customers in finance
Per-tenant AI agents inside a multi-tenant SaaS product
Per-device operators for IoT fleets — homes, vehicles, sensors
Per-property concierge in real-estate or hospitality at scale
Regulated & on-prem agents
Agents in healthcare, finance, and government — running for days in your VPC or on-prem, with audit replay, tenant isolation, and language flexibility.
Examples
Healthcare claim review and medical-necessity decisioning agents
Mortgage underwriting workflow with policy checks and human adjudication
Financial-crime investigation across multi-agent A2A workflows
FedRAMP-, HIPAA-, GDPR-compliant document processing on customer infra
Voice & chat agents
Agents inside real-time channels — voice, chat, and in-product copilots — streaming turn-by-turn while conversations persist across days and redeploys.
Examples
AI receptionist taking inbound calls 24/7 for service businesses
Slack and Discord bots with multi-turn memory across days
In-product voice copilots streaming over WebRTC with low latency
WhatsApp-based booking and customer-engagement agents for SMBs
Classic durable execution
Beyond agents: classic durable execution
Overview
Long-running, multi-step workflows that must not lose or duplicate effects. This is the durable-execution heartland — payments, billing, reconciliation, migrations, cron, and webhook delivery — built on sagas, compensations, exactly-once delivery, and durable timers. It predates the agent era, and on Golem, the same runtime that runs your agents runs it just as well.
Examples
Payment sagas with compensating transactions across PSPs and ledgers
Subscription billing engines with retries, dunning, and proration