Skip to content
Tools & platforms

Every layer has a short list. Here it is.

Vendor-neutral and opinionated. For each layer of an AI system we keep a short list of tools proven in production, what each does, and when we reach for it. Names link to the vendor.

Ask which fits your stack
Tools & platforms

Layer by layer

Grouped by the layer of the system they serve.

Models, sorted by the job they do

We benchmark frontier and open-weight models on your data and route each task to the cheapest one that meets the bar. The frontier moves every few weeks, so the routing layer is built to swap models without touching your product.

“Tag this support ticket”
route
Frontier
Workhorse
Fast & cheap
Open-weight

A narrow step: the smallest model that passes the eval wins.

tap a model to see what it is for

Frontier

The hardest reasoning and long agent runs

Workhorse

Most production agent and chat steps

Fast & cheap

Classify, extract, route and summarise at volume

Open-weight

Self-host for privacy, cost or fine-tuning

Hover or tap a model to see what it is for. We route every request to the cheapest crate that passes your evals.

Agent frameworks & protocols

Standard building blocks for tool use, state, checkpoints and interoperability, so your agents are auditable and portable across models.

LangGraph

Graph and state-machine agent runtime (1.0) with persistence, interrupts and human-in-the-loop; the enterprise default for auditable workflows.
Visit site

Retrieval, search & memory

Vector and keyword indexes, rerankers and memory stores, chosen per corpus size, query type and budget. Postgres-native is the right first choice for most teams.

PostgreSQL + pgvector

Vectors next to your relational data; pgvectorscale for scale.
Visit site

Voice

Speech models, telephony and orchestration for phone and in-app agents, assembled for sub-second turns with semantic turn detection.

ElevenLabs

Low-latency text-to-speech, voice cloning and the ElevenLabs Agents conversational platform (per-minute pricing from about $0.08).
Visit site

Enterprise AI platforms

When you already run on a cloud, a data platform or a CRM, we build inside its AI platform so identity, security and billing stay where your IT team expects.

Amazon Bedrock & AgentCore

Managed models, guardrails and a hosted agent runtime with memory, gateway and identity (AgentCore generally available since October 2025).
Visit site

Data platform

The lakehouse, streaming, transformation and quality layer that turns raw data into trustworthy tables for models and dashboards.

Apache Iceberg

The interoperability standard: hidden partitioning, partition evolution, time travel; native on AWS, Snowflake, BigQuery and Databricks.
Visit site

Observability, evals & guardrails

How we prove quality, catch regressions and keep models inside policy once real traffic arrives. Tracing follows the OpenTelemetry GenAI conventions.

Langfuse

Open-source LLM tracing, evals and prompt management, used across many Fortune 500 companies.
Visit site

Inference, gateways & serving

Where the tokens come from: fast hosted inference, self-hosted serving and the gateway that routes between them. 37% of enterprises already run five or more models in production.

Cerebras

Wafer-scale inference: 2,000+ tokens per second on open models.
Visit site
Tech stack

Languages, frameworks and infrastructure we use every day

The everyday engineering stack under the AI layers.

  • Python
  • TypeScript
  • Node.js
  • Go
  • Rust
  • SQL
  • Bun
  • Swift
  • Kotlin
  • Solidity

Not sure which pieces fit together?

Bring your current stack to a discovery call. We will map the layers and name the gaps.

Book a discovery callSend us a brief