Choose your context
Stop guessing with vector search. Sunder parses your code into abstract syntax trees, instantly mapping every function, its dependencies, and where it's used.
Select your target function, and Sunder retrieves the perfect structural context: the function itself, the children to mock, and the parents to mimic.
Bring Your Own Intelligence
No proprietary lock-in. Sunder runs locally and respects your data privacy using a Bring Your Own Key (BYOK) architecture.
Press [p] to open the model palette. Route specific tasks like baseline coding, adversarial fuzzing, or evaluation to the specialized LLMs of your choice.
Absolute Host Protection
Test with zero risk. Sunder executes untrusted LLM code inside an isolated, ephemeral container built directly from your custom .sunder/Dockerfile.
Host protection is guaranteed through strict zero-trust defaults: your codebase is mounted as read-only, and container network access is completely disabled. Need to test external APIs or heavy workloads? Use the configuration panel to explicitly toggle network access, adjust CPU and memory limits, and inject environment variables.
How Sunder Works
Get Access
Before breaking things, have Sunder write a simple happy-path test to grab valid tokens and database IDs. This prevents the fuzzer from getting blocked by basic auth guardrails.
Attack
Then, perform an adversarial run. Sunder uses the dependencies for perfect mocking and the parents for input and usage mimicry to perform deep logic testing on the target.
Sandbox
The LLM-generated payloads execute inside an ephemeral Docker container, which you specify using a Dockerfile. This code cannot touch your host machine.
Judge and Retry
The Evaluator reads the raw execution logs. If it finds a vulnerability, you get an alert. If the app proves resilient, Sunder learns from the attempt and attacks again.