The Smarter Project
Build AI without engineers
Simple. Fast. In control.
One open-source platform where your people learn to build AI applications, then build the real ones, on infrastructure you own and govern.
- Free and open source
- Runs on your infrastructure
- In teaching use at UBC
Learn. Build. Govern.
Why your organization probably needs Smarter.
Free for instructorsReady-to-teach course designs, from prompt engineering for non-programmers to Kubernetes for platform engineers, with schedules, graded assignments, rubrics and capstone projects.
- course designs
- 21
- micro courses
- 19
- master classes
- 2
- license fees
- $0
- 8 hours classroom bootcamp
- 2 weeks online course
- 4 weeks online course
What your people will buildPhase 2 · Production. These four patterns cover much of what organizations ask for first.
Policy assistant
Answers staff questions from your own handbooks and policies, and cites them.
Student or customer advisor
Lives on your website, under your own name and web address, in your colors.
Data assistant
Answers plain-English questions from approved databases and business systems.
Research assistant
Searches the web and your sources, and shows where its answers came from.
Assistants built on SmarterEach one is a short manifest that ships with Smarter, built from other Smarter resources. Chat with it live, beside the tree of the resources that it is made of.
A weather forecaster
Current conditions and the next hours’ forecast for any place, with a picture of it.
The get_weather tool of OpenAI’s function calling guide is an example that OpenAI never implemented. Smarter did, with one LLMClient that combines four kinds of resources, each a manifest of its own.
- LLMClientweather
The assistant: its instructions, its model and its look.
- Provideropenai
The AI vendor that answers. Its api key is a Secret, encrypted, that neither the model nor the user ever sees.
- Extensionweather
A tool that a developer packaged in Python, and an administrator installed from the signed Smarter Extension registry, without a redeploy.
- ImageSearchPluginimage_search
A tool that is configured, not coded: it finds a picture of the place.
- Guardrailprompt_injection_keyword_input
Screens every prompt for manipulation attempts before the model sees it.
The live chat
Weather LLMClient
Built and run on Smarter
Try asking
- What is the weather in San Francisco?
- I was born December 31, 1966. How many days old am I?
- What courses does Stackademy offer about web development?
Needs a ubc.smarter.sh login. Or run it yourself.
Its architecture
1 moreImageSearchPluginAbout ImageSearchPlugin, in the documentationimage_search1 more
In control
Safe by default, not by policy memo
A policy memo asks people to remember the rules. Smarter enforces them on every message, for every assistant, whoever built it.
- PersonTypes a question
- GuardrailChecks the message
- SmarterUses the stored credential
- AI modelGenerates the answer
- GuardrailChecks the reply
- AnswerDelivered and recorded
Record of every conversation
- 09:14:02ana.r · hr-helpdesk · message received
- 09:14:02guardrail in · nothing found
- 09:14:02smarter · stored credential applied
- 09:14:04model · answer generated
- 09:14:04guardrail out · nothing found
- 09:14:04answer delivered · cost 0.004 USD
- 09:15:31luis.m · sales-quotes · message received
- 09:15:31guardrail in · card number detected
- 09:15:31message blocked · event logged
- Guardrails on every message
Personal data, leaked secrets, manipulation attempts, invented citations and harmful content are checked on the way in and on the way out.
- The AI never holds the passwords
Credentials for your systems are encrypted and stay with the platform. Neither the AI model nor the user ever sees them.
- Spending caps
Budgets in dollars or tokens, per hour, day, week or month, for every person, team and application. Hard stops, not surprise invoices.
- A complete record
Every conversation is recorded, and traceable to the person and budget that allowed it.
Python packages that give your LLMs new tools, installed from the signed Smarter Extension registry without a redeploy.
- 1Browse the registry
- 2Install, signature-checked
- 3Enable, then list it in a manifest
Works with the AI and the platforms you already useCommercial vendors, models you host yourself, and the infrastructure you run
- OpenAI
- Anthropic
- Google Gemini
- Meta
- DeepSeek
- Ollama
- Kubernetes
- AWS
- WordPress
- Drupal
- Shopify
- PostgreSQL
Free and open source, for everyoneSmarter is a community project, not a product. There is nothing to buy: organizations, instructors and students use the same software, under the AGPL-3.0 license.
Free
No license fees or seat counts. You pay AI vendors only for what you use.
Yours
Runs on your own computers or cloud account. Your data stays where you put it.
Portable
Works with OpenAI, Anthropic, Google, Meta, DeepSeek and others, or with models you host.
Open
Every line of code is public, so your security team can inspect it.
Smarter is supported by its community. Organizations that want help installing or running it can work with an independent partner.
The same open source platform, from your laptop to production. No license fees, no seat counts.
1Your laptopDockerQuick Start with Docker Compose
2Your clusterKubernetesThe official Helm chart
3Your AWS cloudTerraformThe official Terraform modules
Answered plainly
Common questions
Cost, data, vendors, support and maturity, answered plainly.
Ask the communitySmarter is free, open-source software that your organization installs and runs itself. People use it to learn how AI works, to build AI assistants that work with your own documents and systems, and to put those assistants in front of staff, students or customers. Leadership uses it to see and control everything AI-related: who built what, what it can reach, what it said, and what it cost.
How to start
Thirty days, with real cases
Days 1 to 7
Platform ready
Smarter is installed. IT sets roles, rules and budgets.
Days 8 to 14
Your team learns by building
A hands-on course for business people. Each one builds a first AI app.
Days 15 to 29
Real cases
The team builds apps for the problems it chose, with company data. Real users try them.
Day 30
Review with leadership
The team shows what it built. Leadership sees real costs and usage, and decides how far to scale.
- Platform running
- Rules and budgets set
- A team that can build
- Real apps working
- Real cost and usage data