# AI SENSE API

> aisenseapi.com is the API host of AI SENSE AS in Oslo, Norway: free public REST endpoints, an MCP server with every endpoint as a tool, and an A2A agent. No account, API key or cookie. The full index of tools, guides and limits is https://aisense.no/llms.txt.

## Start here

- [Full llms.txt](https://aisense.no/llms.txt): every endpoint, tool and guide, with limits and lifetimes.
- [REST API catalog](https://aisense.no/free-public-apis): the endpoints with examples.
- [MCP guide](https://aisense.no/free-public-mcp-server): how to connect and what the tools do.
- [API reference](https://raw.githubusercontent.com/aisenseapi/aisense-free-public-rest-apis/refs/heads/main/API.md)

## Entry points

- REST: https://aisenseapi.com/services/v1/ followed by the endpoint name, for example https://aisenseapi.com/services/v1/uuid. Arguments go in the path, not in a query string.
- MCP: https://aisenseapi.com/mcp, source server version 1.19.0, registry name com.aisenseapi/free-public-tools. The REST endpoints are also MCP tools.
- Audio data: POST /services/v1/audio_encode turns 1 to 128 bytes into a WAV file of datatones. POST /services/v1/audio_decode recovers one checksum-verified message from a PCM16 WAV recording, at most 40 seconds and 8 MiB. Audible is the default, with tones from 800 to 3000 Hz. The experimental high mode uses tones from 14.0 to 16.2 kHz and the experimental ultrasound mode tones from 18.0 to 20.2 kHz. Every symbol hops between three tone sets, so room echo is ignored, and Reed-Solomon repairs damaged bytes. Both high modes depend on the devices and may be audible. Auto detects any of the three bands within one shared time limit. This is unencrypted data transfer, not speech transcription. The MCP tools are encode_audio and decode_audio. See https://aisense.no/free-public-api-audio-encode-api-endpoint and https://aisense.no/free-public-api-audio-decode-api-endpoint
- Decide uses rules by default. The optional models "clef", "nimble" and "tev1" use noul, choice and score questions with instructions and criteria, process the submitted text and have their own per-IP usage limits. A model request may hold 8 KiB, 3 KiB for tev1, the smallest and fastest, tested by its maker in English only and suited to short, direct questions. See https://aisense.no/free-public-api-decide-api-endpoint
- Semantic search keeps short notes in a collection for 24 hours and finds them by meaning, across wording and between languages, with bge-m3 (default) or qwen3-embedding-4b. Answers are ranked suggestions with scores, never a decision. Adding and searching have their own per-IP usage limits. See https://aisense.no/free-public-api-semantic-search-api-endpoint
- One-time Secret: POST /services/v1/secret stores only a locally encrypted AES-256-GCM envelope (16 KiB plaintext maximum) for 24 hours. POST /secret/{id}/reveal consumes before returning ciphertext; GET /secret/{id}/receipt only reads status; POST /secret/{id}/burn revokes. Separate read/receipt bearer tokens belong in Authorization headers. Never send plaintext or decryption keys to MCP or the API; never automatically retry reveal. Creation limit: 100 per IP per UTC day, plus the shared request budget. See https://aisense.no/free-public-api-one-time-secret-api-endpoint
- Agent Handoff: POST /services/v1/handoff keeps one piece of work as a small state machine for 24 hours, with an owner token and a receiver token. POST /handoff/{id}/{state} moves it and names in from the state it expects to leave; any other move is refused with the real state. GET /handoff/{id} reads the state, payload and history, and /handoff/{id}/wait/{seconds} waits for a change.
- Agent Run: POST /services/v1/run keeps a plan of up to 16 steps and goes through them in order for up to 24 hours, also while the client is away: a job for workers, a time, a person or an outside event, Decide rules, and the end. It runs no code and hosts no model. GET /run/{id} reads it, /run/{id}/wait/{seconds} waits for a change, POST /run/{id}/steps/{step}/result hands in a worker's result, and POST /run/{id}/cancel ends it.
- A2A: https://aisenseapi.com/a2a, with the agent card at https://aisenseapi.com/.well-known/agent-card.json

## Limits

- Free allowance: 5000 requests per IP address per day, REST and MCP counted together.
- Optional Pro plans at https://aisense.no/custom-apis share a larger daily allowance across a customer's selected source IPs, without an API key. Endpoint-specific limits and temporary data expiry still apply.
- Stored results and files expire after 24 hours, and each IP address may store 80 MB per day.
- Both day counts 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). A 429 answer carries Retry-After with the seconds until then.
- Anyone who has the link to a stored object can read it, so never store secrets or personal data.