Define custom evals
Track performance of your agents over time
Chef Agent
Retrieval 5.4
Correctness 5.1
Bench Framework
Ship agents that pass the bench
BuildBench is the modern TypeScript framework for building, benchmarking and shipping AI agents — locally first, in production always.
Build and iterate
Agents. Workflows. RAG. Memory. Tools. MCP. BuildBench lets you go from idea to a passing benchmark.
Run your agents in a local dev server
Define agents and workflows in TypeScript. Iterate against the model providers and tools you already use — and watch the bench score climb.
Your developer studio
Visualise traces, run evals and tune prompts in a studio that ships with the framework.
Productionize and test
Tune context. Improve recall. Tweak until your agents achieve human‑level accuracy.
Built-in observability
View traces and logs for your agents
/agent.run
812ms
tools.search_docs
261ms
model.openai.gpt-4o
390ms
tools.send_email
58ms
Deploy and scale
Expose your agents as APIs, or bundle them with your app. With BuildBench, your agents are part of your infrastructure.
Manage deployments
Control your source code and infrastructure
Deployments
#237
#234
● Success
#229
● Success
#221
● Success
Flexible architecture
Deploy BuildBench agents wherever you’re hosting your app, or as a standalone service
Container-based
Serverless
Build your first bench agent
with Templates · read the Bench Book · learn with the Tutorial · watch the Workshop · tune in to Bench Hour
Template
Template
Template
Browser Agent
An agent that controls a real browser to research and act on the open web — scored against a click-trace harness.
Template
Spreadsheet Analyst
Hand a spreadsheet to an agent that summarises trends and exports a tidy report — measured for table-faithfulness.
Template
Chat with Database
A natural-language SQL agent over your warehouse with grounded answers, citations and an automatic recall bench.
Book
Principles of Building AI Benchmarks
A field-tested guide for engineers shipping production agents — and the suites that grade them.
Read the book →
Course
Bench 101 · CLI Course
Learn to wire up agents, tools and MCP servers — and grade them against a real benchmark — in a hands-on CLI course.
Start the course →