Encryption protects what SOL stores. An AI request is different: the material needed for that request must be readable while a service produces a response. This page explains that boundary and names the outside services involved.
Last updated September 20, 2026
The request boundary
Dreams, journal entries, memories, and chat message bodies are sealed before they are stored. SOL does not hold the key needed to open them.
When you use an AI feature, your new input is sent in readable form for processing. If encrypted stored context is included, your device decrypts it first. When memory or semantic search is enabled, SOL can also add relevant past chats, dreams, journals, or memories to the readable request context.
In the standard chat path, user and assistant message bodies are encrypted on your device before database storage. Providers never receive your recovery code or encryption key.
Outside services
One or more services may be involved, depending on the feature you choose. Anthropic produces final responses in Prism Chat by default, plus the shared planning steps. SOL does not automatically send a failed answer to another provider. A documented rollback setting can deliberately move some Prism Chat answers to OpenAI or xAI, but that requires an operator decision.
Read-aloud sends the response text you choose through SOL's servers to ElevenLabs to generate speech. Optional dictation streams interim words to Deepgram from your browser, then sends the final recorded audio clip to OpenAI for the final transcript. The on-device embedding tool loads its reviewed software and model files from SOL's own domain. Decrypted personal text stays inside that local browser worker; the worker cannot fall back to a third-party software or model host.
HTML artifacts are displayed and exported as static, offline documents. Before preview, HTML download, or PDF preparation, SOL removes scripts, forms, external links, and remote resources, then applies a browser policy that blocks scripts and outside connections. Safe text, diagrams, embedded pictures, and native controls such as expandable sections remain available.
AI provider retention boundaries
Anthropic states that standard commercial API inputs and outputs are automatically deleted from its backend within 30 days. Feature-specific terms, safety or Usage Policy enforcement, legal holds, a need to comply with law, and a different customer agreement can change that boundary.
Anthropic also states that commercial API inputs and outputs are not used to train its models by default unless the customer explicitly opts in or submits material through a feedback channel. SOL does not describe this standard boundary as zero data retention, and SOL has not verified an organization-specific zero-data-retention arrangement.
OpenAI states that API data is not used to train its models by default unless a customer opts in. Its default abuse-monitoring logs may retain inputs and outputs for up to 30 days, subject to legal exceptions. If Experience Chat is deliberately rolled back to OpenAI, SOL sends those requests with Responses application-state storage disabled, but that setting is not a zero-data-retention agreement. SOL has not verified an organization-specific OpenAI zero-data-retention control.
Saving encrypted content does not by itself ask an AI provider to process it. Private semantic search and indexing run on your device. If that local search selects relevant private records for a Chat request, the selected readable context is sent to the Chat provider named above when you send the request. Outside-web search through Tavily is disabled for launch; Chat can still search SOL's internal Library. Read-aloud, interim dictation, final dictation transcription, and page-reading actions use the service named above for that path.
Read the full privacy policy