AI Transparency Notice

What Speax is

Speax is a generative AI service. It uses large language models (LLMs) to autonomously plan, reason, and execute tasks on your behalf. You interact with it through natural-language prompts; it responds with text, generated code, research summaries, and other artifacts — some of which may be deployed to the internet.

This means that much of the content you see when using Speax is generated by AI models, not written by humans in real time.

Foundation models used

Speax uses the following third-party foundation models to generate responses:

• Anthropic Claude (Opus, Sonnet variants) — developed by Anthropic, Inc. (USA). Model card
• OpenAI GPT (GPT-4o, GPT-5.x variants) — developed by OpenAI, LLC (USA). Model information
• Moonshot Kimi K2.5 — developed by Beijing Moonshot AI Technology Co., Ltd. (China). Moonshot AI

The model used for a given task depends on your subscription tier and any model selection you make in the interface. All models are accessed via API — Speax does not host these models directly.

What the system does

When you submit a task to Speax, the agent:

• Analyzes your request and breaks it into a sequence of steps using an LLM.
• Autonomously selects and invokes tools — including a web browser, terminal, file system, code interpreter, and external APIs.
• Executes code and shell commands inside an isolated sandbox (Firecracker microVM). The sandbox is destroyed after the session.
• May create files, websites, and other artifacts and, with your instruction, deploy them to public URLs.
• Streams progress updates to you in real time so you can monitor what is happening.
• Terminates the task when it concludes, encounters an error it cannot resolve, or you stop it.

The agent can take actions that have real-world effects: creating files, sending HTTP requests, deploying websites, and (with the right skills enabled) interacting with external services. You should treat Speax's autonomous actions with the same care as you would a person acting on your behalf.

Known limitations

You must not rely on Speax outputs for high-stakes decisions without independent verification. Known limitations include:

• Hallucination: LLMs can generate plausible-sounding but factually incorrect information, including fabricated citations, statistics, and code explanations. Always verify factual claims independently.
• Outdated knowledge: The training data of each model has a cutoff date. Speax may not be aware of recent events, regulations, or technical developments.
• Code errors: Generated code may contain bugs, security vulnerabilities, or logic errors. Review all code before deploying to production or running in sensitive environments.
• Bias: Outputs may reflect biases present in training data. Exercise critical judgment when outputs touch on people, groups, or contested topics.
• Incomplete tasks: The agent may stall, misunderstand complex instructions, or make incorrect assumptions about your intent. Monitoring progress and redirecting when necessary is recommended.
• Irreversible actions: Some actions (file deletion, API calls, deployments) cannot be undone. Confirm before executing irreversible steps.

Human oversight and control

Speax is designed to keep humans in control:

• You can stop any running session at any time using the Stop button. The agent will halt immediately.
• You can observe every action the agent takes via the Operations Panel, which streams all tool invocations and their results in real time.
• You can review and reject outputs before acting on them.
• The agent will ask for clarification when a task is ambiguous rather than proceeding on uncertain assumptions (subject to model behavior).
• Each sandbox session is isolated — the agent cannot access your local machine, other sessions, or other users' data.

Speax is intended as a tool that augments human judgment, not replaces it. Final decisions — particularly those with legal, financial, medical, or safety implications — remain your responsibility.

Model training and your data

We do not train models on your data, and we do not instruct our AI providers to do so.

• Anthropic: the Anthropic API Terms of Service prohibit Anthropic from training on API-submitted data by default.
• OpenAI: the OpenAI API Terms of Service prohibit OpenAI from training on API-submitted data by default.
• Moonshot AI:Speax's agreement with Moonshot AI includes a prohibition on using API-submitted data for model training.

Your prompts and session content may be transmitted to these providers to generate responses. See our Privacy Policy for details.

AI-generated content labeling

Content generated by Speax is created by AI models, not by human authors. When you share AI-generated content publicly — particularly text, images, audio, or code — we encourage you to label it appropriately so that recipients understand its origin.

This is both ethically responsible and increasingly required by law in various jurisdictions (e.g., the EU AI Act requires disclosure of AI-generated synthetic content under certain circumstances).

Contact us about AI concerns

If you have concerns about Speax's AI systems — including potential harms, observed hallucinations, bias in outputs, or misuse — please contact us:

• General: [email protected]
• Safety and abuse: [email protected]
• Security vulnerabilities: [email protected]