VIABLE Open Kit
The VIABLE Harness and VIABLE Engine in one MIT-licensed repository, for developers adding teaching to their own AI and for schools that run classes on their own servers.
About VIABLE Open Kit
The VIABLE Open Kit is the open code of VIABLE: the VIABLE Harness and the VIABLE Engine, with their examples and catalog, in one public repository under the MIT license. It is for developers who want their own agent or education product to teach the way a teacher wants, for people building classroom tools with a coding agent, and for schools, libraries and labs that keep things on their own servers. The VIABLE Learning Platform and the VIABLE Developer Platform run on this same code, and each person takes only the part they need.
The Harness goes around any agent
Before a model replies, the VIABLE Harness adds only what the moment calls for: what the teacher decided, the matching standard, a teaching move and the misconception that fits what the learner did. It also sets the rules, such as how far the AI may help and answer keys it never sees, changes approach when a learner keeps failing, and keeps records tied to what went in.
Four ways into your own AI
A Skill and MCP tools for assistants, a gateway that any app reaches by changing one URL, three functions (prepare, check and record) around your own code, or calls from inside another harness. The gateway, your own calls and the Engine guarantee it runs; a Skill or an MCP tool is advice an assistant may skip.
The Engine runs classes on your server
A school, library or lab gets agents, courses, activities, class lists, learner pages and chat pages, set up in a console in the browser, where Ask can make a course from what a teacher says. Add any OpenAI-compatible model and what it makes possible turns on, from replies to voice and search by meaning. Courses and agents downloaded from the VIABLE Developer Platform open here, and can go back.
Help for the AI that builds
Someone building a classroom tool with a coding agent can give it the building-for-learners Skill, or a block to paste into a chatbot. It has the agent ask who the learners are first, tells it what a class does to a tool, such as thirty learners on one address and a budget per class, and when not to build at all; a review command checks the code. It needs no account.
Checks, tests and an early result
After a reply, rule checks run first, then VETTING, in which a second model reviews the reply against a policy the learner never sees. Agent tests run your own cases or open benchmarks against a version before learners meet it. In an early check on 50 learner messages and two models, a one-line tutor prompt got 8% of first drafts past review and the teacher's settings given before the reply got 68 to 74%, rated by a model, on replies rather than learning.
Defaults that protect minors
No learner's name or identifier is sent to a model, and what is hidden from learners, such as answer keys and a teacher's notes, never reaches their browser. A reply that wasn't checked says so. Running it doesn't by itself make a deployment compliant with COPPA, FERPA or any other law, and the documentation lists what remains to do.
One command opens a real course on your computer, with a panel showing what the AI was given before it wrote a word; it needs no key and keeps nothing, and without a key the model's words are scripted and it says so. Researchers can add a method, a dataset or a benchmark so that teachers' AIs use it with their citation. The Open Kit is version 0.3, a preview made by VIABLE Lab, and until its packages are on npm it is installed from the repository.
At a glance
- VIABLE Harness: what teaching needs, before the reply
- VIABLE Engine: courses and class lists on your server
- Works with any OpenAI-compatible model
- Skills, MCP tools, a gateway or three function calls
- VETTING checks and agent tests
- MIT licensed, in one public repository
Part of the VIABLE family
The lab’s platforms and the classroom tools built alongside them.
- PlatformVIABLE Learning PlatformWhere teachers say what they're teaching and get a course with an AI tutor that helps the way they want, and where learners join it with a link or a code.
- PlatformVIABLE Developer PlatformShared AI infrastructure for education and research: frontier and open-weight models, collaborative agent workflows, knowledge bases, VETTING, and traceable learning experiences.
- Classroom toolCHECKPOINTShort classroom assessments in which every wrong answer points to a misconception, with live sessions students join by code and reports that show teachers who to regroup.
- Classroom toolLOGEN AIA studio where educators write clear, measurable learning objectives with Bloom's taxonomy: AI suggests, a separate AI step reviews, and the educator decides what to keep.
- Classroom toolChronoForgeTurn a lesson into a story learners live: they step into a moment that mattered, investigate, decide under pressure, then compare their path with what really happened.