Launch
Launch

Focus on programming, not prompting.

Focus on programming, not prompting.

Developer tools and cloud infrastructure for perfectionists using LLMs. Zenbase takes care of the hassle of prompt engineering and model selection.

Developer tools and cloud infrastructure for perfectionists using LLMs. Zenbase takes care of the hassle of prompt engineering and model selection.

Mmmmm… code

Scaling
Vibe Checks

Transaction
Categorization

Notification
Relevance

Styled
Generation

RAG Query
Generator

Reliable
Tool Calls

Router with
Latency Constraint

Entity Extraction
with Latency Constraint

It's easier to know what looks good than it is to explain to an LLM how to grade what looks good. Zenbase's classifiers can be calibrated to that of a human reviewer, so that you can programmatically evaluate vibes at scale.

Typescript

Python

// 1. Create your classifier
const vibeCheck = zenbase.classifier({
	id: "vibeCheck",
	params: z.object({
		text: z.string()
	}),
	returns: z.boolean(),
})

// 2. Give it test cases
await vibeCheck.tests.add([
	{ params: { text: "..." }, returns: true },
	{ params: { text: "..." }, returns: false },
])

// 3. Train it
const { evals } = await vibeCheck.optimize()

// 4. Use it
const { data: isVibey } = await vibeCheck({
	text: `A towel, the guide says, is about the most massively useful thing an interstellar hitchhiker can have.`
})

// 5. Incorporate user feedback
const onUserFeedback = (text: string, isVibey: bool) =>
	vibeCheck.tests.add([
		{ params: { text }, returns: isVibey }
	])

Scaling
Vibe Checks

Transaction
Categorization

Notification
Relevance

Styled
Generation

RAG Query
Generator

Reliable
Tool Calls

Router with
Latency Constraint

Entity Extraction
with Latency Constraint

It's easier to know what looks good than it is to explain to an LLM how to grade what looks good. Zenbase's classifiers can be calibrated to that of a human reviewer, so that you can programmatically evaluate vibes at scale.

Typescript

Python

// 1. Create your classifier
const vibeCheck = zenbase.classifier({
	id: "vibeCheck",
	params: z.object({
		text: z.string()
	}),
	returns: z.boolean(),
})

// 2. Give it test cases
await vibeCheck.tests.add([
	{ params: { text: "..." }, returns: true },
	{ params: { text: "..." }, returns: false },
])

// 3. Train it
const { evals } = await vibeCheck.optimize()

// 4. Use it
const { data: isVibey } = await vibeCheck({
	text: `A towel, the guide says, is about the most massively useful thing an interstellar hitchhiker can have.`
})

// 5. Incorporate user feedback
const onUserFeedback = (text: string, isVibey: bool) =>
	vibeCheck.tests.add([
		{ params: { text }, returns: isVibey }
	])

Scaling
Vibe Checks

Transaction
Categorization

Notification
Relevance

Styled
Generation

RAG Query
Generator

Reliable
Tool Calls

Router with
Latency Constraint

Entity Extraction
with Latency Constraint

It's easier to know what looks good than it is to explain to an LLM how to grade what looks good. Zenbase's classifiers can be calibrated to that of a human reviewer, so that you can programmatically evaluate vibes at scale.

Typescript

Python

// 1. Create your classifier
const vibeCheck = zenbase.classifier({
	id: "vibeCheck",
	params: z.object({
		text: z.string()
	}),
	returns: z.boolean(),
})

// 2. Give it test cases
await vibeCheck.tests.add([
	{ params: { text: "..." }, returns: true },
	{ params: { text: "..." }, returns: false },
])

// 3. Train it
const { evals } = await vibeCheck.optimize()

// 4. Use it
const { data: isVibey } = await vibeCheck({
	text: `A towel, the guide says, is about the most massively useful thing an interstellar hitchhiker can have.`
})

// 5. Incorporate user feedback
const onUserFeedback = (text: string, isVibey: bool) =>
	vibeCheck.tests.add([
		{ params: { text }, returns: isVibey }
	])

Is it any good?

We develop custom optimization algorithms and leverage those developed by our Stanford & MIT Research Team @ DSPy.

We develop custom optimization algorithms and leverage those developed by our Stanford & MIT Research Team @ DSPy.

I’ve seen a lot of AI Devtools and Zenbase is solving a problem that everyone building with AI will have when going to production. The best part is their product is so easy to use that it’s a no brainer.

Scott

CEO
Superfilter.ai (YC S24)

CEO
Superfilter.ai
(YC S24)

We were staying up until 3am trying to go from demo to production. Zenbase came into the trenches with us to improve our evals from 10% to 91.6%. It really felt like they were a part of our team.

Taeib

Cofounder
Vera-Health.ai (YC S24)

DSPy's optimizers had 40%+ better accuracy and saved 18 engineer hours on a classification task vs. an expert prompter using 49 prompting techniques.

Learn Prompting

Sponsored by OpenAI & Microsoft

Sponsored by
OpenAI & Microsoft

Feel the Zen

Declarative DX

Zenbase's AI functions let you focus on what you want, not how to prompt or which model to use.

AI that gets smarter all the time

AI that gets smarter with time

We automate the prompt engineering and fine-tuning so you don't have to figure it out.

Never get stuck prompt engineering again

We'll find the best prompt and model to maximize coverage of your test cases.

Your Zen

Masters

Mains

We're two cracked self-taught engineers who've led special projects and teams for 10+ years.

Cyrus Nouroozi

Founder & CEO

• Core Contributor @ DSPy
• Researcher @ Nous Research
• UWaterloo Systems Design Eng

• Core Contributor @ DSPy
• Honorary Researcher @
Nous Research
• UWaterloo Systems Design

Amir Mehr

Founder & CTO

• Core Contributor @ DSPy

• Core Contributor @ DSPy
• M.Sc. of CS @ UCalgary

• M.Sc. of CS @ UCalgary

Supported by angels from

Omar Khattab

Researcher @ Databricks
Creator of DSPy & ColBERT

The Zenbase team has a unique combination of technical expertise and practical problem-solving skills. Their work is both theoretically sound and aimed at solving the real challenges developers face.

The Zenbase team has a unique combination of technical expertise and practical problem-solving skills. Their work is both theoretically sound and aimed at solving the real challenges developers face.

Find your Zen

Open Source

Free Forever

Optimize your existing pipelines

Integrates with your eval tooling

Manual control

Complete control over prompts

MIT License

Startup

Starts at $1000/month

Hosted API

Continual Optimization

Dedicated Slack Channel

Custom Evaluators

Enterprise

Contact Us

SOCII / HIPAA

On-prem deployment

Dedicated Slack Channel

Custom Optimizers

Join the waitlist for our self-serve API

Join the waitlist for our self-serve API

FAQ

You asked, we answered.

How do you find the best prompt & model?

We use AI to continuously experiment with prompts and models to figure out the most effective way to execute your function. We use algorithms developed in-house and those researched by our team at DSPy. Automated prompt engineering combined with expert prompting regularly leads to double digit percentage improvements over expert prompting alone.

Do you offer an on-premise solution / SOCII / HIPAA?

Can we start with a small POC and adopt Zenbase incrementally?

Can I export the prompt & model?

FAQ

You asked, we answered.

How do you find the best prompt & model?

Do you offer an on-premise solution / SOCII / HIPAA?

Can we start with a small POC and adopt Zenbase incrementally?

Can I export the prompt & model?

Made with ❤️ 
in San Francisco 🇺🇸 
and Edmonton 🇨🇦.

Copyright 2024 Zenbase AI Inc.

Made with ❤️ 
in San Francisco 🇺🇸 
and Edmonton 🇨🇦.

Copyright 2024 Zenbase AI Inc.

Made with ❤️ 
in San Francisco 🇺🇸 
and Edmonton 🇨🇦.

Copyright 2024 Zenbase AI Inc.