AI with accountability.

These AI & Automation Terms apply where a florid.io service uses AI agents, generative models, large language models (LLMs), machine learning, retrieval-augmented generation (RAG), autonomous copilots, tool-calling pipelines, or automated workflows.

These terms supplement the general Terms of Use and relevant service agreements. Where a signed enterprise agreement conflicts with these public terms, the signed agreement controls.

AI models generate responses probabilistically and may occasionally output incorrect, incomplete, or biased information.

In florid.io systems, AI models are integrated into robust software architectures alongside business validation rules, deterministic code, audit logs, and human approval checkpoints. The presence of AI does not mean that every action is autonomous or unmonitored.

You retain all rights, title, and ownership in content, prompts, files, and data you provide as inputs to florid.io systems.

You must ensure you have all necessary rights, licenses, and lawful authority to provide inputs, and that doing so does not violate third-party intellectual property or data privacy rights.

Subject to your underlying customer agreement and applicable law, you may use the outputs generated by florid.io systems for your lawful business purposes.

Because generative models operate across shared foundation knowledge, similar prompts submitted by different users may yield comparable outputs. You remain responsible for evaluating outputs before relying on them for mission-critical operations.

Workflows that materially affect individual rights, finances, employment, housing, healthcare, or physical safety must incorporate meaningful human review.

A general-purpose florid.io AI model or agent must never be treated as the sole decision-maker in high-impact scenarios without explicit, bespoke safety designs approved in writing.

Where a customer deploys florid.io intelligent systems to process personal data of individuals in Kenya, the parties must ensure full compliance with Section 35 of the Data Protection Act, 2019 regarding automated decision-making.

Appropriate safeguards must be maintained, including informing data subjects of the automated processing, explaining the underlying logic and anticipated consequences, and providing accessible mechanisms to request human intervention or reconsideration.

Advanced florid.io products (such as Ririkana) utilize long-lived context and organisational memory so agents can remember past decisions, project styles, and business parameters across conversations and teams.

Where organisational memory is active:

• Memory stores are strictly isolated by tenant boundary;
• Role-based access permissions are enforced on retrieved memories;
• Customers retain administrative controls to inspect, correct, or delete memory records; and
• Credentials, secrets, and raw payment numbers must not be submitted into long-term memory stores.

Some florid.io agents may be granted tool-execution capabilities (such as updating CRM records, issuing database queries, creating calendar invites, or dispatching notifications).

Customers configure tool permissions, authorization levels, and spending or execution limits. florid.io designs tool systems with least-privilege principles and confirmation gates for state-altering or high-risk operations.

florid.io utilizes enterprise-grade model providers (such as Anthropic, OpenAI, or Google Cloud Vertex AI) under commercial, zero-retention developer agreements. We do not use consumer-tier AI APIs for customer data.

We implement systematic evaluation suites, output validation checks, prompt grounding, and automated guardrails to mitigate hallucinations, unintended bias, and adversarial attacks. While these measures significantly reduce risks, users must exercise due diligence in high-stakes contexts.

You must not deploy florid.io AI features for:

• Autonomous medical diagnostics or clinical treatment recommendations without qualified physician review;
• Determining criminal guilt, predictive policing, or unlawful social scoring;
• Autonomous denial of credit, insurance, housing, or employment without human review;
• Generating non-consensual deepfakes, impersonations, or deceptive content; or
• Conducting mass automated surveillance or circumventing legal process.

AI model architectures evolve rapidly. We may update underlying model versions to enhance speed, reasoning, or security. Material changes impacting production customer contracts will be communicated with appropriate deprecation windows.

Except where prohibited by Kenyan consumer law, florid.io is not liable for business decisions made on the basis of unverified AI outputs. Clients deploying automated agents are responsible for their configured operational workflows.

If you detect unexpected agent behaviour, hallucinated outputs with safety implications, or potential memory leakage: