Seldon is an applied AI research lab that builds models of human behavior. Its agents are shaped by personality, goals, beliefs, emotions, context, and memory, so they act more like real people in your product. Use Wren for QA testing at $3 per completed run, Finch for enterprise usability research, or Rook for government defense scenarios. Seldon reports its simulated users matched real people's choices 87% of the time.

• Behavioral agents: Each simulated person is modeled on six factors: personality, goals and motivations, beliefs and expectations, emotions and attention, environment and context, and experience and memory. You can change an agent's role or situation to see how its choices shift.
• Plain-language test writing (Wren): Describe a test in a sentence or two, the way you'd explain it to a teammate, including where to start and what success looks like. You can also write full test definitions in YAML.
• Adaptive exploration: Agents choose their next step based on what they see, so a run can surface issues outside your written test cases. Examples include broken layouts, failed API calls, and incorrect totals.
• Simulated usability studies (Finch): Build agents from real clickstream data plus your user context. Then run in-silico usability tests on a prototype, a staging environment, or a live product.
• Journey traceability: Review each simulated journey's actions and decision points to see where an agent understood, hesitated, or got stuck. You can also inspect the goals, knowledge, and context behind each action.
• Ranked insights and recommendations: Finch ranks usability issues, recommends product changes, and shows where different user types behave alike or diverge.
• Calibration against real behavior: Agents are backtested against observed human behavior and refined to better match your users. Seldon reports 87% behavioral fidelity across a full journey.

• Pay-per-run QA pricing: Wren costs $3 per completed test run. Model usage and browser execution are included, with no seat fees or separate token charges.
• Free start for Wren: Each workspace gets its first 10 completed test runs free, so you can try it on real flows before paying.
• No charge for platform failures: Runs that can't finish because of an agent or platform failure aren't billed and don't use your free allowance.
• Less script maintenance: You write tests in plain language instead of scripting every browser action, as you would with tools like Playwright.

• Finch has no self-serve plan: Finch uses custom enterprise pricing with a scoping process and no listed free tier, so small product teams can't try usability simulation on their own.
• Rook is government only: The defense and national security product isn't available to commercial buyers, and its pricing is custom.
• Costs grow with test volume: Every run of a test case in each browser and environment is billed separately, re-runs included. Large regression suites can cost more than flat-rate tools unless you negotiate volume pricing.
• Works best on interactive products: Finch works best with interactive experiences that agents can navigate. Seldon also positions simulated findings as a guide to what to research, with changes still validated with real people.

• Catching checkout bugs before release: An e-commerce QA team describes a flow like "add two items to the bag and check out" in plain language. Wren runs it in parallel and flags issues such as incorrect totals, with recorded evidence and steps to reproduce.
• Finding bugs your test cases miss: An engineering team runs Wren agents that explore adaptively instead of following a fixed script. The agents surface broken layouts or failed API calls that no one wrote a test for.
• Pre-launch usability checks on a prototype: A product team points Finch at a prototype or staging build before running a full user study. They get ranked usability issues and see where simulated users hesitated or got stuck, so live research time goes to the right questions.
• Comparing how different user groups experience a design: A UX researcher builds agents from clickstream data for new and experienced users. Finch shows where their paths and understanding diverge, which helps the team decide which design helps both groups reach their goals.
• Stress-testing product assumptions: A product manager feeds Finch a specific assumption about user behavior. The simulated journeys show whether agents behave as expected, along with recommended product changes to validate with real users.
• Defense scenario modeling: A government agency uses Rook to follow how actors with different goals and information make decisions and affect one another in a wargaming-style scenario.