Tools we build with
A snapshot of the AI models, automation platforms, infrastructure, and developer tools we work with across every engagement. We are loyal to the outcome, not the vendor.
55 tools across 8 categories.
We pick per problem. Frontier model when reasoning is the bottleneck, open weights when cost or residency is. Self-hosted automation when complexity demands it, no-code when speed of launch wins.
Workflow and automation
We build on the platform that fits the team. n8n is our default for self-hosted complexity; Make and Zapier carry the weight when speed-of-launch matters.
n8n
Self-hostable open source automation. Our default for serious workflow builds.
Make
Visual scenarios with strong branching; great for cross-tool integration without code.
Zapier
Fastest path to a working automation when the team has zero engineering capacity.
Pipedream
Code-first automation with built-in event sources; useful for engineer-heavy teams.
Activepieces
Open source automation alternative when self-hosting and licence costs matter.
Temporal
Durable workflow engine for long-running, mission-critical automation.
AI build frameworks
We use frameworks where they earn their place and skip them where direct SDK calls are clearer. The decision is always per project.
LangChain
Composable building blocks for chains and agents; fits some projects, not all.
LlamaIndex
Mature retrieval and indexing primitives; great for document-heavy RAG builds.
Vercel AI SDK
Streaming and UI primitives for AI features in Next.js apps.
Pydantic AI
Structured-output and validation for Python-side LLM calls.
DSPy
Programmatic prompt optimisation; useful when evaluation drives the build.
Haystack
Production retrieval pipelines with strong evaluation tooling.
Infrastructure and hosting
Boring choices when boring works, modern choices when modern unlocks something. We are loyal to fit, not to a stack diagram.
Vercel
Frontend and edge hosting; first choice for Next.js apps and AI surfaces.
Supabase
Postgres + auth + storage; our default backend when speed of build matters.
Railway
Quick container deploys with sensible defaults; great for n8n and worker fleets.
Cloudflare
Edge network, R2 storage, Workers; we lean on it for cost-sensitive performance.
AWS
Used when scale, enterprise integration, or specific services demand it.
Google Cloud
Especially Vertex AI, BigQuery, and the Gemini ecosystem.
Fly.io
Multi-region apps close to users; simple deploys for global services.
Modal
Python-native serverless compute for AI workloads and batch jobs.
Data and retrieval
RAG and analytics live or die on data infrastructure. We pick storage and retrieval primitives that survive the project growing up.
Postgres
Default relational database; usually the right answer when in doubt.
pgvector
Postgres extension for vector search; integrates retrieval with operational data.
Pinecone
Managed vector database when team prefers turnkey retrieval at scale.
Qdrant
Open-source vector database; strong filters and hybrid search support.
Weaviate
Vector + keyword + GraphQL; useful for richer retrieval surfaces.
ClickHouse
Columnar analytics database; great for high-volume event and log workloads.
DuckDB
Embedded analytics; we use it for lightweight ELT and in-process aggregation.
dbt
Transformation layer with versioned, tested SQL; the right way to model warehouse data.
Frontend and product
Frontend choices are about teams as much as tools. Our default Next.js + Tailwind stack scales from internal tools to public-facing products.
Next.js
Default React framework; SSR, edge runtime, and Server Actions out of the box.
React
Component model the rest of the stack is built around.
Tailwind CSS
Utility-first styling; what we ship most production UI in.
shadcn/ui
Composable component primitives; the right base when the design system is being invented.
Framer Motion
Animation library used sparingly to give UIs the right tactile feel.
Astro
Content-heavy sites with islands architecture; useful for marketing surfaces.
CRM and operations
We integrate with the CRM the team already uses. Migration is rarely the right first move; making the existing system smarter usually is.
HubSpot
Most common mid-market CRM in our work; we automate around its API daily.
Salesforce
Enterprise CRM heavy lifter; we build agents and assistants around it.
Pipedrive
Lean sales-team CRM with a clean API; popular with SaaS scale-ups.
Close
Calling-and-CRM combo for high-velocity outbound teams.
Attio
Modern relational CRM with strong API; great for custom workflows.
Intercom
Support and AI assistant surface; we extend it rather than replace it.
Developer experience
The tools that shape how the team writes and ships code. Choices here compound across every project.
Claude Code
Anthropic's CLI for AI-paired engineering; how we move on most builds.
Cursor
AI-first IDE; useful for teams that prefer staying in an editor surface.
GitHub
Source control, CI, code review; the default for every engagement.
Linear
Issue tracking that engineers actually want to use.
Notion
Project documentation and internal knowledge base.
Sentry
Error and performance monitoring across frontend, backend, and AI surfaces.
AI models and providers
We pick models per problem rather than per vendor. Most builds blend two or three providers behind a single internal interface.
Anthropic Claude
Frontier reasoning and tool use; our default for agents and complex work.
OpenAI
GPT and embeddings; broad ecosystem and great structured output support.
Google Gemini
Long-context and multimodal; strong on document and video pipelines.
Mistral
Fast, EU-friendly open and hosted models; useful when residency matters.
Meta Llama
Open weight family used for self-hosted inference and fine-tuning.
Cohere
Command and Embed models, strong for enterprise retrieval and reranking.
Perplexity
Citation-grounded search API for live-web augmented agents.
Replicate
Hosted open-source models for image, audio, and niche workloads.
Together AI
Fast inference for open-weight models, useful for cost-sensitive paths.
Ready to automate your operations?
Walk away with a prioritised list of automation and AI wins, costed, sequenced, and yours. The call is 30 minutes, free, and binds you to nothing. The shortest path to knowing whether AI Workflow Agency is the right fit.