Customer story · Form builder
How Tally red teams the agent that builds its forms
Tally AI turns a plain-language prompt into a working form. Enoki attacks it whenever its tools, system prompt or model change, and re-tests every fix on the next run.
- The agent
- Builds and edits forms from a prompt
- The risk
- Misuse of its form-building tools
- How they test
- On every change to tools, prompt or model
Filip MinevCo-founderEnoki attacks our AI agent the way a real adversary would, then hands us a list of what broke with the exact steps to reproduce it.
Filip MinevCo-founder, TallyAn agent with tools, in front of every user
Tally is a form builder. Its agent, Tally AI, builds and edits forms from a plain-language prompt. That makes it an agent with real tools, acting on the user's behalf down to the styling a form gets.
Filip Minev, co-founder at Tally, works across product and engineering. His team had a security review for the API and evals for the prompts. Neither one tested what happens between them.
If you ship an agent with tools, your risk is not in the model weights. It's in the seam between the model, your tools, your API, and your response stream.
Findings that enter the tracker like any other bug
Tally runs Enoki against Tally AI in production whenever the agent's surface moves: a new tool, a changed system prompt, a new model. Every finding comes back with a replay script, so it goes into their tracker the same way a normal bug does.
The re-runs turned out to matter most. Fixes get verified on the next pass, and some of ours came back marked as regressed. Our own tests would never have told us that.
More customer stories
How Wally tests its accounting agent on every production release
Currently we have built a flow where every time we do a production release, we trigger an Enoki run. This means that Enoki will test all use cases of our AI agent.
Your first assessment
Attack your own agent before an adversary does.
Point Enoki at your agent's endpoint. It attacks it the way Tally's was attacked and shows you the exact steps behind every finding.