OneShot

You're right. Let me be honest about what exists vs. what's designed.

What the project actually does

OneShot is a structured AI agent pipeline. It takes a user intent and runs it through a sequence of stages:

1. Research — an LLM gathers information about the intent and the current repository
2. Plan — an LLM generates a structured plan (goal, tasks, dependencies, validation cases)
3. Audit — the plan is checked against registered rules (dependency cycles, readiness)
4. Build — an LLM generates task-specific execution instructions, then a backend executes them
5. Post-success — documentation is updated based on what changed

Each stage uses a shared generation system (currently Mistral, but model-neutral by design). Rules can be registered and extended. Schemas enforce structural contracts at each boundary.

What exists today

• Researcher, Planner, Builder pipelines — working
• Rule framework with registered rules — working
• Schema validation at input/output boundaries — working
• Iterative plan correction loop — working
• Execution via injected backend — working (interface resolved, concrete backend not supplied)
• Post-success AGENT.md refresh — working

What does NOT exist yet

• Refactor as a distinct step — not implemented
• Validated(id) handoff between audit and build — not implemented
• Expected-vs-actual success verification — not implemented
• Concrete execution backend — not supplied

Why it should exist

The AI agent market today is mostly ad-hoc chains: call a model, parse output, call another model, hope it worked. There's no structural discipline between stages. That makes agents hard to debug, hard to test, hard to extend, and impossible to trust.

OneShot provides the structural skeleton: defined boundaries between responsibilities, schema contracts at each handoff, a rule framework for extensibility, and model-neutral inference so you're not locked to one provider.

The market need is infrastructure for AI agents that teams can actually maintain — not another wrapper around an LLM API, but the engineering layer underneath that makes agents composable, testable, and extensible.

Market fit

• Developer tooling — teams building AI-powered features need structured pipelines, not prompt spaghetti
• Enterprise adoption blocker — enterprises won't deploy agents they can't audit or extend; OneShot provides the structural basis for that
• Model flexibility — as models change (and they will), the inference-neutral layer means you swap the provider without rewriting the pipeline

Built With

• ai-agents
• ai-governance
• built-with:-typescript
• developer
• fastapi
• featherless-ai
• gemma
• google-adk
• json-schema
• litellm
• node.js
• ollama
• opentelemetry
• pydantic
• python
• redis
• rest-api
• sha-256
• sqlalchemy
• sse