TrustInspect

Research / Open Source / TrustInspectOpen-source release planned

Turn AI testing into repeatable evidence.

TrustInspect is a research-driven testing tool for LLM, RAG and agentic applications. It connects test execution with records that help teams investigate behaviour and assess the effect of system changes.

Discuss TrustInspect Explore the approach
Stylised presentation of TrustInspect’s report and terminal interfaces.
Illustrative presentation of TrustInspect. Explore the original interface examples below.
The approach

Connect test conditions, observations and findings.

  1. Define the test context

    Identify the application, the behaviour to examine and the conditions needed to interpret the result.

  2. Capture the evidence

    Record prompts, responses and findings so that observations can be investigated and reported.

  3. Compare and review

    Repeat relevant tests after changes. Use the evidence to discuss improvements, remaining issues and further testing.

Research focus

Testing across modern AI applications.

The research covers these application types; supported integrations and test coverage will be documented with each release.

LLM applications

Prompts and responses

Examine application behaviour under defined test conditions and keep the observations needed for review.

RAG applications

Retrieval and generated answers

Investigate behaviour in the context of connected knowledge sources and the application’s intended access boundaries.

Agentic applications

Interactions and actions

Explore testing approaches for systems that use tools or perform multi-step tasks, with attention to context and observed behaviour.

Continuous testing

Repeat the relevant tests when the system changes.

Define repeatable test conditions so observations from different runs can be interpreted together.
ChangeReview the testing scope
Model or promptsRepeat the behaviours affected by a new model, system instructions or prompt templates.
Knowledge sourcesReview retrieval, access boundaries and responses when connected information changes.
Tools and permissionsRevisit allowed actions and multi-step behaviour when capabilities or permissions change.
Evidence in practice

Support investigation and continuous testing.

Make findings useful to a team.

A result becomes more useful when people can understand the prompt, the response and the circumstances of the test. TrustInspect connects these records to findings for assessment and reporting.

Review what changes over time.

Changes to models, prompts, retrieval sources or system components can affect behaviour. Repeating relevant tests supports comparison and helps teams decide where further investigation is needed.

Inside TrustInspect

From assessment setup to reviewable findings.

These original screenshots show the alpha build supplied for this project, from selecting a target to reviewing the evidence.

01 · Assessment report — view screenshot
TrustInspect alpha report showing an illustrative local assessment and its executive summary.
Example from the alpha build, showing an illustrative local assessment. These figures describe this example run, not overall product performance.
02 · Observations and evidence — view screenshot
Original TrustInspect observations showing test classifications and recorded prompts and responses.
Original report view showing classifications, recorded prompts and responses. Findings require interpretation in the context of the test.
03 · Assessment launcher — view screenshot
Original TrustInspect terminal launcher with target selection and assessment options.
Original terminal view showing the available assessment and target-profiling options.

Follow the project or discuss an application.

TrustInspect’s open-source release is being prepared. Contact us about the research, potential collaboration or testing needs. Repository details and contribution instructions will be added when the release is available.

Contact the TrustInspect team Read our open-source vision