What we test
Safety behavior
Prompt-injection, jailbreak-refusal, personal-data-leak and benign-control test cases, scored the same way every run.
Performance
Latency (p50/p95), throughput, retries and timeouts at a set concurrency.
Agent setups
How an agent is wired: tool permissions, approval steps, secrets handling and what it does with untrusted input such as email.
Regressions
Repeatable runs and a pass/fail threshold you can put in CI, so a drop in quality shows up before release.
Services
LLM Red-Team & Performance Eval Setup
We connect our open-source harness to an endpoint you're authorized to test, build a private test suite, and deliver a first report with safety and performance results.
AI Agent Safety Audit
A fixed-scope, checklist-driven review of an agent setup: injection exposure, tool permissions, secrets and human sign-off.
Monthly Eval Support
Scheduled regression runs and suite upkeep after model, prompt or tool changes.
Built in the open.
Our tools are public and MIT-licensed, so you can read how a test works before it runs against your system.
- llm-redteam-harness (v0.2.0): safety and red-team cases, scoring, latency and throughput reports, and a CI gate.
- gemini-mcp-server: a small MCP server that lets an agent ask Google Gemini for a second opinion, with retries and model fallback.
- cpanel-spam-filter-sync: keeps spam rules in one YAML file and syncs them to cPanel server-side filters, with a dry-run mode.
- agent-email-safety: a design write-up on keeping email-reading agents from following instructions hidden in messages.
Built with AI coding assistance. I defined the requirements, ran the tests, and reviewed the changes. — Leigh Rodney
How we work
Scope
You tell us what the system does and which endpoints we're authorized to test.
Test
We build the suite, run it, and measure safety and performance.
Report and hand off
You get a written report, the test suite, an optional CI gate and the option of ongoing support.
Ground rules
- We test only systems you own or are authorized to test.
- You use your own API keys and your provider's terms apply. We don't resell API access.
- Our work uses AI-assisted tools, and a person reviews every deliverable.
Have an AI system you need to trust?
Tell us what you're building and we'll scope a test plan. Pricing: contact for a quote.