The Infryn guide
Your AI. Your infrastructure. Your control.
From the first credit to the work it powers. Explore the complete Infryn vision, the controls you can use and what comes next.
One platform. The full picture.
The website and product guide are public. Wallet workspaces, access keys, session controls, shared project budgets, encrypted private files, referral tracking, the community leaderboard and activity exports are available in the app. Wallet invitations, Viewer and Developer roles, and limited team keys are available in Projects. Text request access is enabled for eligible models, with streaming, function calls and JSON output where supported. Requests require service capacity, funded API balance and compatible workspace and key privacy settings. Final streamed usage supports automatic settlement recovery after a connection or billing interruption. Nine private and anonymized embedding models are enabled for search, similarity and retrieval applications. Charges use returned input-token counts and the request’s model price. Image generation, stock-voice speech and WAV transcription are available on enabled models. Prices use images, text characters or audio duration, with the same privacy and spending controls. USDG purchase windows open automatically when funded service capacity is available, within the limited pilot on Robinhood Chain. One USDG adds $1 of API balance. No native token is required. Wallet-held credits and the USDG order book are available on Robinhood Chain. Buy and hold, buy and activate, transfer, list, cancel unfilled orders and claim sale proceeds. Orders depend on available liquidity. The INFRYN token is live. Staking will require at least $100 of INFRYN and pause new rewards below that value. Launch awaits an independent price feed, eligibility verification and funded reward periods.
| Capability | Availability | What it covers |
|---|---|---|
| Workspace & access | Available | Wallet sign-in, including deployed contract wallets on Robinhood Chain, workspace settings, scoped expiring API keys, rotation and revocation, privacy preferences, session controls, activity exports and account deletion. |
| Models & developer tools | Catalogue access available | A searchable, paginated Models page spans text, image, video, audio and other model types. Compare privacy, uncensored options, TEE/E2EE capabilities and prices by request type. Text requests support streaming, function calls and structured output on compatible models. Nine private and anonymized embedding models are enabled for search, similarity and retrieval applications. Charges use returned input-token counts and the request’s model price. Image generation, stock-voice speech and WAV transcription are available on enabled models. Prices use images, text characters or audio duration, with the same privacy and spending controls. Video, music, image editing and encrypted clients need further integration. Account endpoints, usage exports and scoped MCP account tools are available. Text request access is enabled for eligible models, with streaming, function calls and JSON output where supported. Requests require service capacity, funded API balance and compatible workspace and key privacy settings. Final streamed usage supports automatic settlement recovery after a connection or billing interruption. |
| Text embeddings | Available | Nine private and anonymized embedding models are enabled for search, similarity and retrieval applications. Charges use returned input-token counts and the request’s model price. Send up to 16 text inputs per request. Workspace, key and shared-project limits apply. Input text and returned vectors are not stored in workspace records. Private files are not sent automatically. |
| Image & audio APIs | Available on enabled models | Image generation, stock-voice speech and WAV transcription are available on enabled models. Prices use images, text characters or audio duration, with the same privacy and spending controls. Generate one 512px or 1024px image, synthesize up to 4,096 characters with a catalogue voice, or transcribe a PCM WAV file up to 60 seconds and 3 MiB. Input and generated content are not stored in workspace records. |
| USDG API balance | Limited pilot | USDG purchase windows open automatically when funded service capacity is available, within the limited pilot on Robinhood Chain. One USDG adds $1 of API balance. No native token is required. Purchase history, saved transaction references, settlement recovery and CSV exports are included. |
| AI credits | Available | Wallet-held credits and the USDG order book are available on Robinhood Chain. Buy and hold, buy and activate, transfer, list, cancel unfilled orders and claim sale proceeds. Orders depend on available liquidity. One credit activates into $1 of API balance. Activation is irreversible. |
| Staking & rewards | Independent price feed pending | The INFRYN token is live. Staking will require at least $100 of INFRYN and pause new rewards below that value. Launch awaits an independent price feed, eligibility verification and funded reward periods. Claim to your wallet, activate as API balance or list at your chosen price. Earned rewards remain claimable after withdrawal. Signed automatic-listing policies use a five-credit threshold and require onchain execution. |
| Credit market | Available | Wallet-held credits and the USDG order book are available on Robinhood Chain. Buy and hold, buy and activate, transfer, list, cancel unfilled orders and claim sale proceeds. Orders depend on available liquidity. Quotes specify maximum USDG and minimum credits. No market or activation fee; network gas is separate. Optional swaps, card purchases and native-token proceeds conversion remain on the roadmap. |
| Agents & automation | Access controls available | Keys have permissions, expiry and spending caps. MCP tools check balances, models, request status, credit activity and staking. A separate permission prepares unsigned credit and staking transactions for wallet review. Bounded replenishment remains on the roadmap. Reward listing policies are implemented; network deployment and a running executor are required. |
| Anonymized access | Available on eligible models | Conceals account identity from the model provider. The provider still sees prompt content and applies its own data policies. Both workspace and key must explicitly permit this level; it cannot satisfy Protected or Confidential requirements. |
| Protected access | Available | Text request access is enabled for eligible models, with streaming, function calls and JSON output where supported. Requests require service capacity, funded API balance and compatible workspace and key privacy settings. Final streamed usage supports automatic settlement recovery after a connection or billing interruption. Requests are readable by the service and model provider. Requests cannot fall back to weaker privacy. |
| Confidential access: TEE & E2EE | Roadmap | TEE-capable models support hardware-isolated execution. E2EE-capable models support client-side encryption to protected compute. Capability labels are visible; verified Confidential execution and end-to-end encrypted requests are not active in Infryn. Unavailable requirements are refused. |
| Encrypted private files | Available | Browser-encrypted files and filenames, protected by a passphrase that stays on your device. Up to 20 files, 1 MiB per file and about 10 MiB total. Search supported text files on your device, with matching excerpts and line references. Search phrases and readable content are not sent to the service. Download and decrypt locally; deletion removes the active copy. File sizes, counts and times remain visible. Infryn cannot recover the passphrase. Files are not sent to AI models. |
| Projects & shared budgets | Available | Group API keys under one lifetime inference spending limit. Pending and settled usage count across all assigned keys, including rotated or revoked keys. Pause new requests or adjust the limit. Key limits and available balance still apply; no automatic budget resets. |
| Project teams | Available | Wallet invitations, Viewer and Developer roles, and limited team keys are available in Projects. Invite a wallet for seven days. Viewers read the project budget; Developers create limited inference keys. Up to ten members and pending invitations per project. Keys share the project limit and owner privacy requirement. Removal or changing to Viewer revokes their keys. Personal balances, private files and account settings stay separate. |
| Private workspaces & workloads | Roadmap | Model-connected retrieval, private agent jobs and dedicated endpoints. Local text search in encrypted files is available separately. |
| Privacy beyond content | Research | Reduce exposure of connection and payment information. No promise of anonymous network access or anonymous public wallet activity. |
| Usage & participation | Available | Private request reporting, activity exports, referral links and an automatic alias leaderboard. Referral commissions are 1.5% of eligible settled purchases, with a 14-day hold and $10 minimum payout. Eligible pilot purchases accrue commissions. |
Last reviewed 2026-09-25. Availability applies to the specific capability shown, not every feature in its category.
Credits that connect participation and AI.
The credit economy connects participants who earn funded rewards with builders who need AI access. Stake to participate in allocations, claim credits, keep what you need and transfer or sell the rest. Buyers can acquire credits without staking.
The native token, transferable service credits and activated usage balance are separate. Activating credits converts them into balance for supported AI requests. It does not create withdrawable cash. Rewards depend on available funding; no fixed yield, redemption floor or guaranteed return is promised.
USDG purchase windows open automatically when funded service capacity is available, within the limited pilot on Robinhood Chain. One USDG adds $1 of API balance. No native token is required. Wallet-held credits and the USDG order book are available on Robinhood Chain. Buy and hold, buy and activate, transfer, list, cancel unfilled orders and claim sale proceeds. Orders depend on available liquidity. The INFRYN token is live. Staking will require at least $100 of INFRYN and pause new rewards below that value. Launch awaits an independent price feed, eligibility verification and funded reward periods. One credit activates into $1 of API balance.
The planned staking rule requires at least $100 of INFRYN to earn new rewards, pausing new rewards below that value until topped up. The historical eligibility contract is implemented and tested; launch still awaits a verified independent token price feed, network deployment and funded reward periods. Full withdrawals and already earned rewards must remain available without a working price feed. Rewards use your stake over time within completed, funded hourly periods. Unfunded hours promise no rewards, and earned credits remain claimable after withdrawal. Claim to your wallet, activate them or list them for USDG.
Optional automatic listing requires a signed price policy. Once at least five earned credits are ready, a separate onchain transaction can list them at that price. You remain the seller and control cancellation and proceeds. Disable future listings at any time; existing orders remain open until filled or cancelled. A running executor is required for unattended listing. Public wallet activity includes stakes, rewards and orders.
Purchase windows accept 5–100 USDG, with a 100 USDG per-account daily limit and capacity checks before you send. Review the exact network, asset and recipient in the app. Network gas is paid separately. Submitted transaction references are saved to your account; final settlement adds balance once. Reopen a purchase to check a pending transfer or recover its reference. Never resend merely because settlement is pending. Purchase and usage histories can be exported as CSV.
API balance is for usage, not cash withdrawal. Unspent-balance refunds require review and a confirmed return transfer; contact our official social account with your purchase reference. Never share an API key or seed phrase. Wrong-network, wrong-asset and out-of-window transfers need manual review and are not automatically credited.
A market for the credits you need.
The credit market supports buying and holding credits, buying and activating in one action, priced sell listings, partial fills, cancellation of unsold credits and USDG proceeds claims. Review maximum payment, minimum credits, the recipient and gas in your wallet before signing. Quotes do not reserve liquidity. Unspent payment stays in your wallet. No market or activation fee is charged.
Optional token swaps, card purchases and conversion of sale proceeds extend the roadmap. Each depends on supported payment routes and available liquidity. Wallet-held credits and the USDG order book are available on Robinhood Chain. Buy and hold, buy and activate, transfer, list, cancel unfilled orders and claim sale proceeds. Orders depend on available liquidity. Direct USDG API-balance top-ups are a separate flow.
After a wallet transaction, wait for final settlement. Reopen the market to check its transaction hash or recover wallet history. The same activation is credited only once. Transfers, sell orders and proceeds claims are public wallet activity; privacy settings for AI requests do not hide them. Buying from another user pays that seller and does not earn a referral commission. Never send tokens directly to a contract address.
AI access where you already work.
Infryn centers on developer tools, applications and agents. Wallet workspaces, access keys, session controls, shared project budgets, encrypted private files, referral tracking, the community leaderboard and activity exports are available in the app. Text request access is enabled for eligible models, with streaming, function calls and JSON output where supported. Requests require service capacity, funded API balance and compatible workspace and key privacy settings. Final streamed usage supports automatic settlement recovery after a connection or billing interruption.
Keys reveal their secret once. Rotation replaces a key while preserving its expiry, permissions and remaining spending budget. Pending requests must settle first. Permissions distinguish account reads, inference and unsigned wallet preparation. Team keys use narrower model-read and own-request-status permissions, without account-wide access. A spending cap limits authorization; it does not fund an account. Model availability, prices, privacy eligibility and supported features can differ.
The app provides an API connection example and an MCP endpoint for scoped account tools. Read balance, models, request status, credit activity and staking. The separate transactions:prepare permission returns unsigned credit or staking transactions, including exact approvals. Your wallet must review and sign them; tools cannot broadcast or take spending authority. Contract deployment and availability checks still apply. Key spending caps apply to inference, not wallet transactions. Use a separate key for each client and keep it in secure local configuration. The roadmap adds broader client integrations.
Chat completions support text, streaming, function-call messages and JSON response formats on compatible models. Clients execute their own tools; Infryn does not run them.
Set a unique Idempotency-Key for each new request and check its status after an interrupted stream. When a finished response includes validated final usage, settlement can recover automatically without running the request again. If final usage is missing, its reserved balance stays held for evidence-based review; an interrupted connection alone does not prove that no charge occurred. Image generation, stock-voice speech and WAV transcription are available on enabled models. Prices use images, text characters or audio duration, with the same privacy and spending controls. File inputs, vision, video, music and image editing need further integration.
Text embeddings.
Nine private and anonymized embedding models are enabled for search, similarity and retrieval applications. Charges use returned input-token counts and the request’s model price. Use POST /v1/embeddings with model: "text-embedding-bge-m3", a text input or an array of texts, and optional encoding_format: "float" or "base64". Use an inference-enabled key and a unique Idempotency-Key. Responses include vectors and input-token usage.
Limits are 16 inputs and 32,768 UTF-8 bytes of text per request. Each input is limited to the lower of 8,192 bytes and the model’s input-token limit. The Models catalogue identifies private and anonymized routes; workspace and key privacy must both permit the selected model. Models that support custom dimensions accept an optional integer dimensions up to their native vector size. Embedding requests use your key, project and account limits. The input price is fixed when the request is admitted; completed usage is billed by returned input tokens, with charges rounded up to the balance’s smallest unit. Input text and vectors are not stored in workspace records. Your client stores and searches its own vectors. Private files stay separate and are never submitted automatically.
Interrupted requests.
Check GET /v1/requests/your-idempotency-key before retrying. If complete billing evidence was received but settlement failed, it can be recovered without repeating the request. A timeout before that evidence arrives remains held for review; an incomplete request is not automatically charged or treated as free.
Give agents useful access, with boundaries.
Start with separate keys, explicit permissions, expiry and spending limits. These controls are implemented. Read-only account tools can check a balance, list connected models and inspect a request made with the same key. Bounded credit replenishment and optional automatic reward listings remain roadmap features.
Purchase and trading authority remain separate from inference access. An agent should not gain permission to top up, trade or change privacy settings just because it can make a request. Each automation needs revocable permission and its own limits. An agent with an owner-issued inference-write key can submit an already-signed activation transaction to POST /v1/credits/confirm with {"transactionHash":"0x…"}. Team-issued keys cannot use this endpoint. This only confirms settled credits for the key’s own wallet; it cannot authorize a purchase, transfer or trade.
Images, speech and transcription.
Image generation, stock-voice speech and WAV transcription are available on enabled models. Prices use images, text characters or audio duration, with the same privacy and spending controls. Use models marked enabled in the app or GET /v1/models. Image prices are per image, speech prices per million input characters and transcription prices per audio second. Final prices are shown before use. These requests share your key permissions, workspace privacy, project limits and available API balance.
Generate an image.
Send POST /v1/images/generations with model, prompt, n: 1 and size of 512x512 or 1024x1024. Prompt length depends on the model. The JSON response contains one data[0].b64_json image. Image editing, upscaling and remote image URLs are not accepted.
Turn text into speech.
Send POST /v1/audio/speech with model, input, voice and response_format. Choose a stock voice and an available MP3 or WAV format from the model catalogue. Input is limited to 4,096 characters, counted as JavaScript string length. Optional speed ranges from 0.25 to 4. The response is audio bytes. Cloned voices and voice URLs are not accepted.
Transcribe a recording.
Send multipart data to POST /v1/audio/transcriptions with model and a file. Use uncompressed PCM WAV, mono or stereo, up to 60 seconds and 3 MiB. Supported sample rates are 8, 16, 22.05, 24, 32, 44.1 and 48 kHz. Duration is measured from the validated audio samples. The response is {"text":"…"}. Optional WAV recorder metadata and the original filename are removed before submission; spoken information remains part of the audio.
Use Authorization: Bearer $INFRYN_API_KEY and a unique Idempotency-Key for every new request. Completed requests return X-Infryn-Cost in USD and X-Infryn-Request-Id. Outputs are returned directly and are not saved in the workspace. Interrupted or unverifiable responses hold their reservation for review; check request status before trying again. These APIs process readable content under the selected model’s privacy boundary and do not provide Confidential execution.
Choose the protection your work needs.
| Option | Purpose | Current boundary |
|---|---|---|
| Anonymized | Shield account identity from the model provider | The provider sees prompt content and applies its own data policies. No blanket no-retention or no-training promise. Both workspace and key must permit this level. |
| Protected / Private | Require limits on data use and content retention | Uses eligible private routes. The service and model provider process readable requests. This is not end-to-end encryption. |
| TEE | Hardware-isolated model execution | TEE means Trusted Execution Environment. Catalogue badges identify capable models; Infryn's verified Confidential execution is still on the roadmap. |
| E2EE | Encrypt prompts on the client for protected compute | E2EE means end-to-end encryption. Catalogue badges identify capable models; client encryption and verification are not active in Infryn yet. |
Text request access is enabled for eligible models, with streaming, function calls and JSON output where supported. Requests require service capacity, funded API balance and compatible workspace and key privacy settings. Final streamed usage supports automatic settlement recovery after a connection or billing interruption. The Models page lets you search and filter by model type, privacy, uncensored options, TEE and E2EE support. It distinguishes model capabilities from available Infryn integrations. Eligible text models support streaming, function calls and structured output. Nine private and anonymized embedding models are enabled for search, similarity and retrieval applications. Charges use returned input-token counts and the request’s model price. Image generation, stock-voice speech and WAV transcription are available on enabled models. Prices use images, text characters or audio duration, with the same privacy and spending controls. Video, music and image-editing endpoints remain pending.
Workspace and key preferences set independent minimum privacy requirements. Existing preferences do not change when new models appear. Selecting Anonymized for a key cannot override a Protected workspace, and Confidential cannot be satisfied by an identity-shielded or plaintext request. Unsupported requirements are refused without a weaker fallback.
Private execution records, reduced connection-data exposure and reduced linkage between payments and AI work extend the vision. These are not current anonymity guarantees. Public wallet activity remains public, and connected tools have their own data practices.
The work around AI belongs here too.
Personal workspace settings, content-free activity exports and private usage CSV exports are available. Projects group API keys under a shared lifetime spending limit. Wallet invitations, Viewer and Developer roles, and limited team keys are available in Projects.
Projects and shared limits.
Create up to 20 projects in your workspace and choose a project when making a key. Each request must fit the project limit, the key limit and available balance. Settled and pending usage count across all assigned keys. Rotation, expiry and revocation do not reset project usage. The assignment cannot be changed after creation.
Pause a project to stop new requests; existing requests can finish. Limits have no automatic reset and cannot be lowered below spent and pending usage. Project limits cover inference, not wallet transactions.
Project teams.
In Projects, the owner can invite a wallet as a Viewer or Developer. Invitations appear when that wallet signs in and expire after seven days. Viewers can read the shared budget. Developers can create up to three active inference keys, subject to the owner’s workspace limit of twenty. Each project supports ten active members and pending invitations; an account can join up to fifty projects.
The owner funds the project and controls its limit and minimum privacy. Team keys allow inference, model discovery and their own request status. They cannot read the owner’s account balance, use account-wide agent tools or prepare wallet transactions. Removing a member, leaving a project or changing a Developer to Viewer revokes that member’s keys. Requests already running may finish. Owners can also revoke individual team keys. Review membership before sharing a budget.
Private files.
Private files are encrypted with their filenames in your browser before upload. Choose a unique passphrase of at least 16 characters and keep it in a password manager. Infryn cannot recover it.
Storage locks when you hide or leave the page, or after 15 minutes without interaction. Limits are 20 files, 1 MiB per file and about 10 MiB total, subject to available service capacity. Downloaded files are decrypted locally. File counts, encrypted sizes and upload times remain visible to the service; a secure browser and device remain necessary. Files are not sent to AI models. Deleting a file removes the active copy; encrypted backup copies follow service retention. Resetting storage deletes all active files and lets you choose a new passphrase.
Find text in private files.
Unlock storage and search for a phrase across all your stored TXT, Markdown, CSV, TSV, JSON, JSONL and log files. Search is case-insensitive and shows up to three excerpts per matching file with line references. Results use five files per page. Download the file to read every match. PDFs, images, binary files and unreadable text are skipped and counted.
Search and decryption happen on your device. The service receives requests for encrypted files, but no search phrase or readable file content. Nothing is sent to AI models. Locking storage, hiding the page or leaving Private files cancels the search and clears its view. Clear search removes the results without locking storage. Model-connected retrieval, private agent jobs and dedicated endpoints remain roadmap capabilities.
Aggregate economic reporting will distinguish funding, outstanding obligations and usage. Referral tracking and the community leaderboard are available in the app. Every account appears under an alias. Wallet addresses, spending and AI activity are excluded from the public leaderboard.
Invite builders. Share in what follows.
Every account has a referral link and a public alias. A valid referral can be replaced before the referred account’s first settled purchase; attribution then stays attached. Self-referrals and circular referrals do not qualify.
Referrals earn 1.5% of eligible settled purchases, including repeat top-ups. Commissions have a 14-day hold and a $10 minimum payout. Refunds reduce commissions proportionally; disputed payments remain on hold. Payouts go to the referring account’s wallet in USDG on Robinhood Chain after review and confirmed transfer. Eligible settled pilot purchases accrue commissions under these terms.
The leaderboard includes all accounts automatically. Each first qualifying referral earns one point. Clicks and account creation earn no points. Points have no cash value and do not promise a token allocation. Choose an alias that does not identify you if you want to keep that public profile separate.
Before you start.
- Check the capability table before relying on a workflow. Staking deployment is pending. Wallet-held credits and USDG API balance are separate; check liquidity and your privacy requirements before buying.
- Compute requires funding. “Unlimited possibilities” describes the vision, not free or unlimited compute.
- Model availability, privacy support and client compatibility vary.
- Prices, fees, eligibility and transaction terms must be available before purchase or participation.
- No published security audit, performance benchmark or service-level guarantee is claimed.
Your data, with clear boundaries.
This notice covers the public website and explains account data handling. Wallet workspaces, access keys, session controls, shared project budgets, encrypted private files, referral tracking, the community leaderboard and activity exports are available in the app. No advertising analytics or session replay is installed.
Website visits.
The landing page and guide have no prompt field or payment form. Fonts, artwork and video are served from this site. Visiting these pages does not send requests to an AI model or external video service.
Connection information.
The hosting service receives an IP address and request metadata when you visit. A specific retention period for hosting records has not been verified. Infryn does not claim anonymous network access.
Wallet connections and accounts.
Connecting a wallet uses third-party wallet discovery, connection and verification services. They may receive connection or device metadata. Sign-in asks for an authentication signature, not a transaction or spending approval. Public wallet addresses and onchain activity are not private. Deployed contract wallets using standard message-signature verification can sign in while connected to Robinhood Chain. Wallets that have not been deployed and contract accounts on other networks are not supported.
The account system stores a wallet-linked workspace, preferences, project names and budget settings, access-key records and administrative events. All accounts have a public alias and participation score. Your private wallet identity, balance and inference activity are not published in that profile. Referral attribution links accounts internally for eligibility and commission accounting. Session cookies expire after 12 hours; sign-in challenges expire after five minutes. Signing out invalidates that session. Settings lists active sessions and lets you end other sessions without collecting device fingerprints or IP addresses for that list. Key secrets are shown only when created and are not stored in readable form.
Project collaboration.
Invitations store the invited wallet, role, expiry and membership state. The project owner can see invited wallet addresses, roles and project-key usage, including key names, limits, expiry and billing metadata. Members see the project name, owner alias, budget totals and assigned key count; Developers also see their own project keys. Other members’ wallets, personal files, payment history and account settings are not shared. Leave shared projects before deleting your account. Revoked memberships and key records may remain with the owner for access and usage accounting.
Contract-wallet sessions.
For contract wallets, we retain the signed sign-in proof for the session and check its authorization on authenticated workspace requests. Network verification services receive the contract address, message hash and signature. A rejected signature ends that session; an unavailable check blocks access until it can be verified. API keys are separate delegated credentials and remain valid until revoked, expired or limited. Review and revoke existing keys when wallet ownership changes.
Requests and records.
The application does not store prompts or responses in workspace activity. Billing records may contain request identifiers, model identifiers, token counts, image counts, speech character counts, audio sample counts and rates, request prices, processing time, amounts and settlement state. Embedding input text, returned vectors, image prompts, generated images, speech text, audio and transcripts are not stored in those records. Payment and referral records include wallet addresses, transaction references, attribution, commissions and payout state. We keep these records separate from prompt and response content. Private and anonymized processing expose readable content to the service and model provider. Anonymized routes may have different downstream retention and training policies. Confidential protection, including verified TEE execution and end-to-end encryption, is not currently available.
Private files.
Files and their filenames are encrypted in your browser before storage. Infryn stores the encrypted copies, a password-derived verification value, file counts, encrypted sizes and upload timestamps. The passphrase is not sent to the service. A strong unique passphrase and a trusted browser/device are necessary; a weak passphrase can be guessed against encrypted data. Infryn cannot reset a passphrase while preserving those files. Private files are separate from inference and are not sent to models. Text search downloads encrypted files and searches locally; phrases, readable filenames and excerpts remain in the browser while storage is unlocked. Locking storage cancels the search and clears its view. Downloaded copies are decrypted on your device.
Retention and control.
Links to our social account open a third-party service under its own data practices. No social tracking widget is embedded.
Revoke keys and sign out to end that access. Workspace records persist; signing out does not delete an account. Deleting a workspace removes its active account records, keys, sessions and public profile. Accounts with financial records require review before deletion so balances and obligations remain accounted for. Public wallet history remains onchain. Hosting and database backups may retain copies outside the active application; their retention periods have not been verified. The activity view covers the past 90 days; older activity and expired sign-in records are removed by bounded cleanup jobs. Financial records remain for accounting and dispute handling. Service checks use aggregate costs, balances and error states without prompt or response content. No prompt history is kept in the workspace. Do not submit sensitive work to an unavailable capability.
Updated 2026-09-25.