AI Model Provider Policy

For this policy, "Avalon," "we," "us," or "our" means Avalon Flow Inc., a subsidiary of Questili LLP, unless a signed order form or customer agreement identifies a different contracting entity.

This policy explains how Avalon handles hosted AI providers, BYO keys, local models, private models, self-hosted inference, custom endpoints, training/no-training commitments, and fallback behavior.

1. AI features covered

Avalon may use AI systems for summaries, drafts, classifications, semantic search, due-date suggestions, meeting briefs, Daily Pulse, AI memory, workflow planning, analytics, suspicious/cold email filtering, connector assistance, and automation support.

2. Hosted AI providers

Avalon may use hosted AI providers such as OpenAI, Anthropic, Google AI, Groq, OpenRouter, Vercel AI Gateway, AWS Bedrock, Perplexity, or similar providers depending on configuration.

When hosted AI features are used, Avalon may send relevant Customer Content, prompts, metadata, and workflow context to the selected provider to deliver the requested feature. Avalon aims to send only the context reasonably necessary for that feature.

3. Customer-selected providers and BYO keys

Customers may configure BYO model providers, private endpoints, API keys, model gateways, or customer-selected routes where supported. Customers are responsible for provider approval, credentials, data-use terms, logging, retention, model behavior, security, and compliance for those routes.

Avalon does not guarantee the quality, safety, legality, availability, retention behavior, or security of customer-selected providers unless Avalon expressly agrees otherwise in writing.

4. Local models and self-hosted inference

Local models, private models, open-source models, self-hosted inference, and customer-controlled custom endpoints are customer-controlled systems. Customers are responsible for licensing, hosting, securing, monitoring, validating, and maintaining those systems.

Customers must accept and comply with the Local Model / Customer-Controlled AI Risk Acknowledgement before using those routes where required.

5. Training and product improvement

Avalon does not sell Customer Content for advertising. Avalon does not use Customer Content to train public third-party foundation models unless that use is explicitly disclosed and authorized through the applicable product setting or customer agreement.

Customer-selected providers, BYO providers, local models, custom endpoints, and self-hosted systems may have their own data-use, logging, retention, and training behavior. Customers are responsible for reviewing and accepting those terms before routing Customer Content to them.

6. Prompt and output minimization

Avalon should send only the context reasonably necessary for the enabled feature. Normal telemetry, analytics, logs, and support bundles should avoid raw Customer Content, prompts containing Customer Content, model completions containing Customer Content, AI memory text, secrets, OAuth tokens, API keys, passwords, and payment data.

7. Output quality and human review

AI outputs may be inaccurate, incomplete, outdated, unsafe, biased, or unsuitable for a specific use. Users must review outputs before relying on them or using them externally.

Avalon is not a substitute for professional judgment and must not be used as the sole basis for legal, compliance, employment, financial, medical, safety-critical, or other regulated decisions.

8. Fallbacks and provider changes

Avalon may change, disable, or route between AI providers, models, or endpoints for security, reliability, cost, quality, availability, provider policy, abuse prevention, or customer configuration reasons.

Fallback behavior may change output style, latency, cost, feature availability, quality, or supported context length. Customers with strict model-routing requirements should configure approved providers and disable unwanted routes where supported.

9. Prompt extraction and model distillation

Users may not use Avalon, hosted models, local models, BYO providers, custom endpoints, logs, traces, exports, scraping, browser automation, or repeated querying to extract, derive, reconstruct, distill, copy, train on, fine-tune on, benchmark against, or otherwise attempt to learn Avalon’s non-public prompts, system instructions, internal policies, model-routing logic, safety rules, tool schemas, evaluation data, source code, hidden workflows, or proprietary operating methods.

10. Contact

For AI provider, model-routing, or local-model questions, contact support@avalonflow.com.