Code-aware analysis
Parses each component with ts-morph to understand props, hooks, context usage, and rendered output.
AI-powered developer tooling
Unit tests,
generated where
your code lives
testgen is a local-first AI test generation tool. It reads your React components through static analysis and produces ready-to-run React Testing Library suites - entirely on-premises, with zero source code leaving your environment.
100% local inference
VS Code + CLI
RTL + Jest
Zero data egress
1
Parse
ts-morph extracts props, hooks, and JSX structure into a component profile.
2
Generate
A LoRA-tuned local model writes an RTL suite from the profile via Ollama.
3
AccountsSummary.test.tsx
renders balance for the user
expect(screen.getByText(...))
Suite passes - reviewed by developer
100%Local inference, zero egress
4Step pipeline to test suite
75%+Coverage uplift target
Phase 3Pilot validation, in progress
Overview
Built for teams that can't send code outside
Fully local, zero egress. The AST and the TypeScript compiler drive generation — no source code, no user data, no proprietary logic ever leaves your environment. A local model via Ollama is optional and only engages when the compiler cannot resolve a failure.
Optional local model
An Ollama-hosted model runs on local hardware and is only invoked when the deterministic pass cannot resolve a failure — never by default.
Zero data egress
Every stage - parsing, prompting, generation - executes on-premises, built for strict compliance environments.
How it works
From component to test suite in four steps
Static analysis does the heavy lifting up front, so the model's job is focused: write meaningful assertions for behavior already mapped by the analyzer.
Parse
The target file is loaded into a ts-morph project. Props, exports, hooks, and JSX structure are extracted into a structured component profile.
Contextualize
The profile is enriched with repo conventions - test file placement, naming patterns, and the providers a component renders under.
Generate
The fine-tuned model produces a React Testing Library suite from the profile, covering render paths, interactions, and edge states.
Deliver
The suite is written alongside the component, ready for the developer to run, review, and commit - developer always stays in the loop.
CLI / VS Code
Developer targets a component file or folder - single file or batch.
ts-morph AST
Props, hooks, context, and JSX extracted into a component profile.
Ollama - LoRA model
Fine-tuned DeepSeek-Coder writes assertions from the profile - 100% local.
Test suite written
Delivered next to the component - developer reviews, runs, commits.
Dual-provider architecture - Ollama runs local by default; the provider layer is pluggable, so a future remote model can slot in without touching the CLI or the analysis stage.
Features
What you get
CLI-first generation
Run testgen against any component file or folder. Batch mode generates suites across an entire feature set in one command.
Dual-provider architecture
A configurable provider layer supports local Ollama inference by default, with a pluggable design for evolving model strategy.
Coverage uplift
Targets lifting unit test coverage toward the 75%+ range, without writing boilerplate by hand.
Reusable skill packaging
Repo-specific conventions are captured in a portable testgen.skill package, benefiting every team that adopts the tool.
42%baseline75%+target
Typical pilot repo before adoption vs. the target range testgen is built to close, without hand-written boilerplate.
Manual~45 min
testgen~5 min
Per-component effort to reach a reviewed, passing test suite - writing assertions by hand vs. generating and reviewing.
Usage
Quick start
# Single component $ testgen generate src/components/Button.tsx # Batch - entire feature folder $ testgen generate src/features --recursive # Output written next to each component ✓ Button.test.tsx generated ✓ Modal.test.tsx generated ✓ Form.test.tsx generated
The CLI writes test files next to their components, following each repository's existing naming and folder conventions.
| Layer | Technology | Runs where |
|---|---|---|
| Static analysis | ts-morph (TypeScript AST) | Local |
| Generation model | DeepSeek-Coder-1.3B - LoRA via Ollama | Local |
| Test framework | Jest + React Testing Library | Local |
| Interface | CLI | Local |
Roadmap
Development phases
testgen follows a phased delivery model - each phase validated on real repositories before the next begins. We are currently in Phase 3.
Foundation & AST engine
Built the core static analysis pipeline using ts-morph to extract component structure into a reliable intermediate profile. CLI interface established.
LoRA fine-tune & local inference
Fine-tuned DeepSeek-Coder-1.3B via LoRA on curated codebase examples. Integrated Ollama for on-premises inference. Dual-provider architecture built for flexibility.
Pilot validation across repositories
testgen has produced passing test suites across a growing set of pilot repositories. Each run tightens provider detection, environment setup, and convention alignment.
Layered orchestrator
A decision-tree orchestration layer that routes each component through the optimal analysis depth and prompting approach, automatically.
CI/CD integration & coverage dashboards
Pipeline hooks that detect coverage gaps and propose candidate tests during pull requests, plus per-repo dashboards tracking uplift over time.