Our tooling
The stack.
The exact tools we use to ship production agent workflows. Transparent by design: few in our market publish their stack, and we think that's a mistake.
Why we publish this.
Two reasons. First, it lets you evaluate us before the call. Second, it proves principle four: we build for portability. Everything here is open-source, self-hostable, or industry-standard with portable data. If we go away tomorrow, you hire someone who knows the same tools.
Agent orchestration
LangGraph: stateful multi-agent workflows
Pydantic AI: type-safe agent logic
LangChain: when the ecosystem matters
LLM providers
Anthropic Claude: primary for complex reasoning
OpenAI GPT-4o: fallback and vision tasks
Google Gemini: long context, multimodal
Observability
Langfuse: traces, evals, cost tracking
OpenTelemetry: structured telemetry
Background jobs
Trigger.dev: durable workflows, scheduling
Celery: Python task queues
Data layer
PostgreSQL: primary store, pgvector for retrieval
ClickHouse: event analytics at scale
Qdrant: dedicated vector search
Infrastructure
Vercel / Railway: app hosting
GCP / AWS: cloud compute
Docker + GitHub Actions: CI/CD