Stacks/Fintech / internal platform

Block KGoose (Money Bot, Manager Bot, and G2)

Block's Applied AI team built KGoose as one Temporal-based agent harness behind several customer and internal products. A ZenML write-up of Catherine Zhang's talk describes the move from synchronous chat to async Temporal workflows after long tool-call loops caused timeouts, plus DynamoDB-backed history, message compaction, external guardrails, and scale of 100 million-plus weekly activities and 10,000-plus concurrent workflows.

Use case

A shared AI agent platform at Block that powers Money Bot (a Cash App financial assistant for peer-to-peer payments), Manager Bot (a Square merchant assistant), and G2 (an internal productivity platform)

Category

Fintech / internal platform

Team size

Built by a small team (exact size not stated)

Orchestrator / framework

KGoose evolved from a synchronous one-shot query API, to a synchronous stateful chat API, to an asynchronous Temporal-workflow harness. Clients submit input and poll for output. Temporal was already widely used at Block for over 100 services, including tier-zero payments and the Square ledger. The platform processes over 100 million weekly activities, with 10,000+ concurrent workflows running at any time.

Memory

Version 2 introduced DynamoDB for tracking chats and messages, moving message-history persistence onto the service. Context limits are handled with message compaction. Redis is listed among the external systems the platform relies on.

Evals

The write-up notes the team was 'not really at the point where we're optimizing agent performance yet' at the time of the talk.

Observability

not stated as a named product. Operational concerns discussed include non-determinism (sometimes throwing away workflow executions), cancellation, and compliance.

Monthly cost

not stated

Models

not stated

Observability tools

not stated

Other tools

DynamoDBRedisExternal guardrail servicesGoose (open-source agent, Rust)

Source

https://www.zenml.io/llmops-database/building-production-ai-agents-at-scale-with-temporal-and-kgoose

Catherine Zhang (Block Applied AI), as documented in the ZenML LLMOps Database · 2026