Privacy
What our free public REST APIs record, how long it stays, and who else sees it. Written from the service source rather than from a template, so it describes this system and not a generic one.
Last updated 9 November 2026.
Who operates this service
The service is operated by AI SENSE AS, Postboks 1202 Vika, 0110 Oslo, Norway. Org.nr NO 922 601 151 MVA. Questions about this page go to support@aisense.no.
The short version
The free APIs do not require account registration or payment details. Content sent through temporary endpoints can still include personal data. Agent Inbox stores sender addresses and cleaned mail content, including any names or addresses in that content. The public utility endpoints set no cookies. The Pro customer portal uses essential security cookies. We run no browser analytics, advertising or tracking scripts. What we do record is the technical detail of each request, because a public service has to be able to see its own traffic.
What we record
Every request
One line in a server log, written when the answer has been sent: the time the request arrived, your IP address, the HTTP method, the requested path with supported bearer credential segments redacted, the protocol, the status code we answered with, the content type, your browser or client's User-Agent string, and how long the answer took.
URL inputs can still enter the log. Bearer IDs for Storage, Capture, Approval, Schedule, Agent Wake, Heartbeat and Inbox are redacted in supported paths. Queue credentials belong in the Authorization header and Queue query strings are not logged. Other query strings and ordinary URL inputs can be recorded. Several endpoints take their input in the URL rather than in a request body - /url_shortener/{url} receives an entire target address, /ip_reverse_lookup/{ip} an IP address, /domain_ip_lookup/{domain} a domain name. Those values are written to the log along with the rest of the line. Do not put anything sensitive in a URL, here or anywhere else.
MCP request summaries
The MCP method log records time, IP address, method, tool name and User-Agent. It does not record tool arguments or results. Daily summaries retain counts by tool and User-Agent, plus a capped list of individual tool-call events containing time, IP address, tool and User-Agent. These are operational traffic records, not an anonymous count.
Rate limiting
We count requests per IP address to enforce the free limit of 5000 per day. The counter is stored against the address. Since 12 September 2026 we also count the bytes each IP address stores through /storage, to enforce the limit of 80 MB per day. That counter is stored against the address in the same way and resets at the same time.
Optional decision models
The default rule mode of /decide computes the answer in the API itself and does not save request or answer bodies. Selecting a model such as Clef sends the state and questions to the model service, which processes them to produce the answer. The model service receives only the state and questions, not your IP address or other request details. The API records usage counters keyed by an IP HMAC and aggregate durations, outcomes, bytes and token counts. It does not save model input or answer bodies. Ordinary access logs still record IP addresses as described above.
Do not send credentials or sensitive personal data in model requests.
Semantic search
/semantic_search keeps the notes you add, with their optional keys, until the collection expires 24 hours after creation, together with the vectors the embedding model makes from them. Adding notes sends their text to the model service, and a search sends the search text, which is not stored. The model service receives only that text, not your IP address or other request details. The API records usage counters keyed by an IP HMAC and aggregate durations, outcomes, bytes and token counts for these model calls. Deleting a note removes its text and vector at once.
Anyone holding a collection's read token can search its notes and read their text. Do not add credentials or sensitive personal data to notes.
Data over sound
/audio_encode and /audio_decode process the submitted data or recording for that request and do not store the audio or decoded payload. The receiver page asks for microphone access when you start recording. It submits the recording to the API when you stop and decode, or when the 40-second recording limit is reached. You can cancel before submission. The sound is not encrypted, and nearby devices can record it. Ordinary request logging still applies.
The MCP decoder accepts a file already in Storage. If you upload a recording there, the separate Storage lifetime of up to 24 hours applies. The audio tools do not extend that lifetime.
What temporary endpoints store
The following endpoint groups hold temporary state or content:
| Endpoint | What is stored |
|---|---|
/storage | Your request body, byte for byte, unchanged |
/url_shortener | The address you asked us to shorten |
/webhook_capture | The complete inbound request, including all headers except the value of X-AISENSE-Key, and the sender's IP |
/webhook_action | The form you defined and the answer someone submitted to it |
/webhook_schedule | The target URL, delivery time, JSON payload and delivery result |
/agent_wake | Task state and the webhook, human answer or time result. Standard credential headers on webhooks are redacted |
/heartbeat | Check-in timing, the webhook URL and optional payload, or an Agent Wake task ID. The target is removed when the monitor becomes terminal |
/inbox | Disposable address, hashed inbox credential, sender address, subject, received time, cleaned message text, extracted codes and public links. Attachments and raw MIME are not kept in the API state |
/queue | Jobs, payload content, status, attempts, timing, deduplication keys and hashes of role tokens and claim receipts |
/semantic_search | Note text, optional keys, note IDs and timestamps, the embedding vectors of the notes and hashes of the read and write tokens. Search text is not stored |
/lease | Hashed lease identity and owner data, timing, fencing token and an optional completed result. Raw namespaces, keys, owner tokens and fingerprints are not stored |
Webhook Capture deserves particular attention. It records every header a caller sends, and services that post to webhooks routinely include authorisation tokens and signatures in those headers. Webhook Capture replaces the value of X-AISENSE-Key with [redacted] before storage. Other headers, including Authorization, are recorded as received. If you point a third-party service at a capture URL, its credentials will sit in our storage for 24 hours. That is the endpoint working as intended, but you should know it before you use it.
Lease replaces values under secret-shaped field names such as authorization, password, token and api_key with [redacted] before storage. It cannot detect a secret placed under an ordinary field name. Keep credentials and personal data out of Lease results and Heartbeat payloads.
The Queue service writes payload content to its JSON state files without encrypting it. It normalizes object-key ordering and JSON serialization. Authorized readers and claiming workers can read the content. Keep credentials and sensitive personal data out of them. Queue processing does not execute payloads, fetch URLs, make callbacks or send content to outside services.
Agent Queue uses separate read, write and worker bearer tokens issued only at queue creation. A queue ID alone grants no access. REST clients send the required token in the Authorization header, never in a URL. MCP clients pass it as a tool argument. A claim receipt identifies the current attempt and is disclosed only to the claiming worker.
One-time Secret
The One-time Secret browser tool encrypts text on the device using AES-256-GCM and a freshly generated key before upload. The API receives an encrypted envelope, not the plaintext or encryption key. The key and read token are carried in the share URL fragment, after #, which browsers do not send in HTTP requests. A separate receipt link contains only the ID and receipt token.
The service stores the encrypted envelope, hashes of the read and receipt tokens, and lifecycle metadata. Access ends after one delivery, revocation or a fixed 24-hour expiry. Encrypted records and metadata may remain until cleanup, including across service restarts. This is not a guarantee of secure erasure from disks, backups, memory or recipients' devices. Operational request logging remains subject to the logging and retention practices described on this page.
The tool page uses no browser analytics, third-party scripts, remote fonts, local storage or session storage. It removes the URL fragment from the address bar after its script starts and keeps access details only in page memory. The device, browser extensions, page code, clipboard and sharing channel can still access the message or complete link. Recipients can keep copies. A delivery receipt records that the service released encrypted data; it is not evidence of identity, decryption or human reading.
REST and MCP clients must encrypt and decrypt locally. Never send plaintext, encryption keys or a complete share link to an MCP tool or an AI conversation. MCP clients may retain operation tokens in their own logs and conversation history. Read the One-time Secret documentation for the access model.
Agent Run
Agent Run stores the plan the creator sends, the results handed in and the answers to its waits, hashes of the read and control tokens, the tokens of the queue, handoffs and Wake tasks it makes, and the creator's IP address, which charges a run's calls to that address and is never answered. A cloud provider in the EU runs this service for us and holds these until the run's own deadline of at most 24 hours, and the expired record is removed 48 hours after creation. Daily counts per address hold only a hash of the address and are removed after three days. Keep credentials and personal data out of plans, tasks and results. Plans and results are stored separately from the encrypted credentials used for managed execution.
For a managed run, you install your OpenAI or Ollama Cloud API key through a private HTTPS request. For optional delivery, the same request provides a Telegram bot token and chat ID or one Slack or Discord incoming webhook URL. We encrypt these details for that run using an environment-specific key. The executor can decrypt them while it works. It removes its usable credential record when it processes completion, failure, cancellation or expiry. You can revoke the key or webhook at its provider.
Your chosen model provider receives the task and the context used to answer it, which can include excerpts from authorized sources and tool results. We fetch the public HTTPS addresses you grant. The source receives the request for that URL. If you enable delivery, the chosen messaging service receives the message and any links it contains. Anyone with a shared form or Storage link can use it until its own expiry. Copies held by external providers follow their retention rules.
A task and the result before it also travel as a job in Agent Queue, and a person's or an outside event's answer arrives through Agent Wake. Those services keep them on our own server for their own lifetimes, as described on this page, also after a shorter run has ended. Rules go to Decide, and a delegated step is shown in Agent Handoff with only the run's and the step's IDs. The application access log keeps the path, whose ID is not a credential, and drops any query string. Read the Agent Run documentation for the access model.
Agent Handoff
Agent Handoff stores the JSON payload the creator sends, the notes written with each move, the declared states and allowed moves, hashes of the owner and receiver tokens, the history of moves with the time of each and which token made it, and, when a notify URL is given, that URL and the creator's IP address for the delivery worker. Nothing is encrypted at rest and both tokens read the payload and the notes, so keep credentials and personal data out of them.
Access ends at the fixed 24-hour expiry and an hourly cleanup removes the files. A notify signal carries the handoff ID, the states and the time of the move, never the payload, a note or a token. The application access log keeps the path, whose ID is not a credential, and drops any query string. Read the Agent Handoff documentation for the access model.
What we do not do
No cookies from the public utility endpoints. The customer portal uses essential security cookies. No advertising, browser tracking scripts or automated decisions about access based on personal profiles. Operational request statistics are described above. We do not sell or rent anything to anyone. Public utility requests do not require an account. Pro customer records are described below.
IP geolocation via /ip_reverse_lookup runs against database files on our own server. The address you look up is not sent anywhere.
How long we keep it
| Data | Kept for |
|---|---|
| Storage, URL shortener, webhook capture, webhook action, Agent Wake | Up to 24 hours |
| One-time Secret encrypted envelopes and receipt metadata | Message access ends after one delivery, revocation or the fixed 24-hour expiry. Encrypted records and metadata can remain until cleanup, including across restarts |
| Managed Agent Run model key and optional delivery credentials | Encrypted per run. The executor removes the usable record when it processes a terminal state. Messages and shared links have their own lifetimes. |
| Agent Run plan, results and creator's address | Until the run's fixed deadline of at most 24 hours, and the expired record is removed 48 hours after creation. Nothing extends it |
| Agent Handoff payload, notes and history | Until the fixed expiry 24 hours after creation. Moves do not extend it |
| Webhook Schedule | Scheduled for up to 24 hours. The final result can remain readable for another 24 hours |
| Heartbeat | Armed for no more than 24 hours. Terminal timing and delivery state can remain readable for another 24 hours |
| Lease | Until the fixed absolute expiry 24 hours after the first acquisition. Renewals do not extend it |
| Agent Inbox API state | At most 24 hours from creation. Activity does not extend expiry |
| Temporary DNS names | 24 hours from creation, or until deleted. The name and its address are public in DNS for as long as the name lives. The creator's address, a hash of the token and the timestamps are kept with the record on both name servers and are removed within a minute of expiry. The lookup counters on the primary count queries per name and hold no client addresses |
| Raw Inbox mail at the receiving provider | Separate mailbox and provider retention. The 24-hour API expiry does not delete this copy, and the default worker does not configure a purge period |
| Agent Queue, including completed and failed jobs | Until the fixed expiry 24 hours after queue creation. Activity does not extend it |
| Semantic search collections, notes and vectors | Until the fixed expiry 24 hours after creation. Activity does not extend it. A deleted note's text and vector are removed at once |
| Rate limit counters | Shared request counters and the storage byte counters reset at 22:00 UTC from late March to late October and at 23:00 UTC the rest of the year (midnight in Norway, time zone Europe/Oslo). Queue and semantic search creation use separate fixed 24-hour windows. Inbox creation uses a UTC-day quota |
| Raw server access and MCP method logs | Temporary local copies are pruned after 14 days. Compressed archives are kept for approximately 24 months, with the cleanup threshold set to 744 days |
| MCP daily summaries | No automatic expiry is configured for the daily summary files |
| Machine-access audit log | 12 weeks |
| Decision-model and semantic search telemetry | Snapshots show up to seven UTC dates. Older aggregates and daily IP-HMAC counters are removed on the next state write. With no activity, older physical data can remain. No model bodies or answers are saved in these records |
| Client login attempts | The email address and the time stay in the internal messaging service our team uses until we delete them there. The counters that limit repeated attempts hold a hash of the IP address, not the address, and are kept for one hour |
Once the stated readable period ends, the data becomes unreachable through the API. Cleanup runs on a schedule, so removal of the underlying file can happen after the stated API expiry.
Where data goes
Some endpoints contact another service or a destination you selected. Only the value needed for that call is sent:
| If you call | We ask | And send |
|---|---|---|
/bitcoin/balance/{address} | blockchain.info | The wallet address |
/solana/balance/{address} | api.mainnet-beta.solana.com | The wallet address |
/ethereum/balance/{address} | ethereum-rpc.publicnode.com | The wallet address |
/webhook_schedule | The public webhook URL you supplied | Your JSON payload and the schedule ID |
Capture or Approval with notify_url | The public callback URL you supplied | A completion signal and result URL. The signal does not include captured content or answers |
/inbox | The configured Gmail mailbox over IMAP | Incoming raw mail is received by the provider and read into cleaned API state. The default worker marks messages as seen without deleting them. Permanent deletion of imported mail is optional |
/heartbeat with a webhook action | The public webhook URL you supplied | The heartbeat miss event and your optional JSON payload |
/blake3_hash, /hash_verify with BLAKE3, /argon2id_hash, /bcrypt_hash, and /password_verify for Argon2id and bcrypt strings | Cloudflare, which computes these hashes for us | The bytes to hash, or the password and the hash to verify. Not your IP address or the headers of your request |
/audio_render | Cloudflare, which makes the sound for us | The notes, or the MIDI file or module. Not your IP address or the headers of your request |
/run | A cloud provider in the EU that runs this service for us | Your plan, the results and answers it receives, and your IP address |
/webhook_schedule, a notify_url and a Heartbeat webhook | A cloud provider that makes these calls for us | The URL you supplied and what the call sends. Not your IP address |
Creating a short link stores the destination mapping within this service and returns a 307.fi URL. Creation does not contact the destination or send it to a separate redirect provider.
For the hash endpoints above this includes the password you send, so use test data with them and hash production passwords inside your own application. Apart from those, /run and the calls to a URL you supplied, no free REST endpoint sends your supplied content to a fixed third party, and an Agent Wake heartbeat action stays inside AI SENSE.
The sign-in page links to the Pro customer portal on aisenseapi.com. Credentials are submitted there over HTTPS. Passwords are verified against a stored hash and are not forwarded to our internal messaging service.
Calls from other websites
Our API responses permit cross-origin requests from any website, which is what makes the endpoints usable directly from browser JavaScript. A consequence worth stating plainly: if a site you visit calls our API from your browser, your IP address reaches us and is logged, even though you never visited us. We have no way to know which site that was, and we do not receive a referrer for it.
Your rights
Under the GDPR you may request access to personal data we hold about you, its correction or erasure, a restriction on how we use it, or a copy of it - and you may object to our use of it. Write to support@aisense.no.
One practical limit, stated because it is true rather than to avoid the obligation: log entries and rate-limit counters are stored by IP address and nothing else. If you ask us to find your records, an IP address is all we can search on, and shared or dynamic addresses mean we may not be able to distinguish your requests from someone else's. Data you stored through /storage is different - quote the storage_id and we can act on it precisely, though it will have expired within 24 hours anyway.
You may complain to the Norwegian Data Protection Authority, Datatilsynet, at datatilsynet.no.
Why we are allowed to do this
We rely on legitimate interest for the technical logging and rate limiting described above: operating a public service, keeping it available, and investigating abuse and faults. Pro account information is used to provide the account and manage the agreement you request. The data is limited to what those purposes need. Where you send us content deliberately - anything you post to the storage or webhook endpoints - you are asking us to hold it, and we hold it only for as long as stated.
Security
All traffic is served over HTTPS. The optional Pro service stores password hashes, hashed session tokens and hashed API tokens in private customer storage. Temporary data is addressed by an unguessable bearer identifier, or by a Lease namespace and key. Treat these values as secrets. Anyone holding the right value can read the associated public state or result.
Changes
When the service changes, this page changes with it. The date at the top is the last time it was checked against what the code actually does.
Pro customer accounts
The optional Pro service stores your name or company name, email address, password hash, selected IP networks, which start with the address you register from, the API tokens you make, as a hash of each with its label, last four characters and the times it was made and expires, plan, orders, discounts and daily usage counts. It uses secure, essential cookies for sign-in and form protection. The public API still works without an account.
Customer records persist beyond the 24-hour lifetime of temporary API data. They are kept to manage access and billing. Email addresses are not verified automatically. Contact support for account recovery or deletion. When Stripe checkout is enabled and you choose it, we send Stripe your account name, email, order reference, plan and discount. Stripe collects payment and billing details on its own pages. We store Stripe customer, checkout, subscription and invoice references, payment status and the paid-through date to manage your allowance. AI SENSE does not collect full card numbers or card security codes. Stripe describes its handling of personal data in its privacy policy.
The Pro service also keeps a log of account events for 90 days: registrations, sign-ins, orders, the steps to and back from Stripe checkout, plans a confirmed payment made active or that ended, and forms it refused, with the message shown and the form field it was about. An entry holds the time, the account and the plan, whether a discount code was entered, and for a new account or a confirmed payment the account name. It never holds a password, a discount code or an email address, and an entry for a refused registration holds nothing that identifies the person. AI SENSE staff read the log to see where customers get stuck. When an account is created or a payment makes a plan active, its name and plan are posted to the internal messaging service our team uses.
The registration form has a short check against automated sign-ups: you click three symbols in a picture in the order asked, and meanwhile your browser solves a small computing task, which takes about a second. The check sets no cookie and sends nothing to a third party. The task is kept for at most 24 hours with no address in it, the points you click are checked once and not kept, and for a new account the event log keeps how long the computing task took, rounded to a tenth of a second. Opening the form counts toward a daily limit for your address, kept as a SHA-256 hash of the address until the next day's count replaces it.