MatrAIx¶

Development line: project:matraix · thread persona-8b-public-documentation
Last event: 2026-08-11 · 1 dated since 2026-08-11 · Researched: 2026-09-05 · confidence: high

What it is¶

MatrAIx runs LLM agents with user profiles across Survey, AI Chatbot, Web, and App tasks.

• 1 290 categorical persona attributes.
• Public Persona 1M with 599 847 grounded and 400 000 synthetic records.
• 1 010 tasks across more than 25 domains.

We use it to generate and stress-test hypotheses, not to replace research with real people. It fits reproducible product and agent evals as long as results are not treated as human validation.

Development line¶

• 2026-08-11 — MatrAIx Persona 8B was publicly referenced with a visual playground quickstart. The practical scope covers Survey, AI Chatbot, Web, and OS-app. Real runs require model provider API keys. Web and OS-app require Docker.

What changed¶

2026-08-04 — MatrAIx published its technical report. Persona 8B claims 8.3 billion records, a Playground with four environments, and 18 189 completed eval trajectories. 2026-08-11 — MatrAIx was available as an open Playground and task library. The practical scope covers Survey, AI Chatbot, Web, and OS-app. Real runs require model provider API keys, and Docker is required for Web and OS-app. Note on the 2026-08-11 step: the initial release is MatrAIx-Persona-8B, not a standalone Persona-8B inference model. Its public dataset is Persona 1M, not the full 8.3 billion claimed personas. New dated steps: 2026-07-31 opened the Playground and task library; 2026-08-01 published Persona 1M; 2026-08-04 released the report on arXiv.

How to use this¶

As of 2026-08-11, practitioners should begin evaluation or hands-on use of MatrAIx Persona 8B from its repository and visual-playground quickstart.

1. Install Python 3.12, uv, and Docker. On Windows, use WSL2 because native PowerShell and cmd do not support task verifiers. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B
2. Clone the repository, create a uv environment, and install the core package with Playground, Rewardkit, and Harbor-LangSmith. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B
3. Verify the environment with a Survey and Chat smoke test without Docker, then run Harbor smoke for Web and OS-app with Docker. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B
4. For real evals, set a compatible provider API key, pick a task and persona, generate a job recipe, and run it with matraix run. Get results with matraix results. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B/blob/main/docs/quickstart.md#10-playground--play-tasks-visually
5. For a cohort rather than a smoke test, download Persona 1M and set it as the dataset in the Playground or CLI. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B

Best practices¶

• Start with dev-sample and smoke tests to verify the setup, not to draw conclusions about user populations. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B
• For research, use Persona 1M instead of the local sample of roughly 200 personas. Pin the recipe, model, and cohort so the run is reproducible. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B
• Treat simulation as hypothesis generation and stress testing, then confirm significant product decisions with real user data. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B
• Do not plan Web and OS-app evals without Docker. On Windows, keep the repository on the WSL2 filesystem rather than /mnt/c. — https://github.com/MatrAIx-ai/MatrAIx-Persona-8B

Superseded by this¶

• Nothing marked obsolete yet.

Still unknown¶

• Official sources show no separate release dated exactly 2026-08-11. The step reflects the documentation and repository state available on that date, not a formal release.
• No independent practical validation confirms that simulation results predict user behavior in a specific product.
• The repository develops without GitHub Releases. No release version pins the compatibility of a specific commit with future provider APIs.

Sources¶

Agent brief¶

• Subject: project:matraix, thread persona-8b-public-documentation, 1 dated events 2026-08-11 → 2026-08-11.
• Practical note: As of 2026-08-11, practitioners should begin evaluation or hands-on use of MatrAIx Persona 8B from its repository and visual-playground quickstart.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.