aisense-free-public-rest-apis

# Free Public REST APIs - AI SENSE AS

No API key | No sign-up | No cost

Provided by **[AI SENSE AS](https://aisenseapi.com)** (Oslo, Norway).
Full endpoint reference: [`API.md`](API.md) | Repo: [github.com/aisenseapi/aisense-free-public-rest-apis](https://github.com/aisenseapi/aisense-free-public-rest-apis)

**Base URL:** `https://aisenseapi.com/services/v1/`

## Free public MCP endpoints

Production discovery on 9 September 2026 returned 28 AI SENSE workflow tools at:

`https://aisenseapi.com/mcp`

The AI SENSE MCP server covers Heartbeat, Lease, Agent Wake tasks, Agent
Inbox, human approval, webhook capture, temporary storage, URL shortening,
time and UUIDs. It needs no account or API key. Heartbeat uses
`create_heartbeat`, `read_heartbeat` and `ping_heartbeat`. Lease uses
`create_lease_namespace`, `acquire_lease`, `renew_lease`, `release_lease` and
`complete_lease`. Agent Inbox uses `create_agent_inbox` and
`read_agent_inbox`. See
[`MCP.md`](MCP.md) for the tool list, data boundary and client examples.

The eight Agent Queue tools are included in that discovery result. The deployed
Queue REST smoke test passed 21 checks on 9 September 2026. Queue operations
use separate read, write and worker tokens issued at creation.

Start with [AGENT-GUIDE.md](AGENT-GUIDE.md) to choose tools, then
[AGENT-QUICKSTART.md](AGENT-QUICKSTART.md) for a complete Queue workflow and
retry decisions.

The official MCP Registry lists `com.aisenseapi/free-public-tools` version
`1.8.0` as active, verified on 9 September 2026.

Verifyum has its own dedicated MCP endpoint at
`https://api.verifyum.com/mcp`. It exposes the three Verifyum proof operations
without an account or API key. File hashing still happens on the agent's
machine. The browser flow, public HTTP API and published protocol remain
available. The official MCP registry lists it as `com.verifyum/mcp` version
`0.1.0`. Finalized proofs also join hourly and daily Merkle checkpoints in
the [Verifyum Witness Layer](https://verifyum.com/witness). Nine records
surround each finalized proof.

| Tier | Records |
| --- | --- |
| Primary evidence | One finalized Solana Mainnet Memo transaction per proof |
| Independent corroboration | Hourly OpenTimestamps on Bitcoin, daily qualified EU timestamp, daily witness-cosigned Sigsum and daily Certificate Transparency certificate |
| Operator records and availability redundancy | Verifyum Ed25519 signature, GitHub checkpoint log, Software Heritage and Internet Archive |

The Solana transaction is the primary evidence. Deep Solana history generally
requires an archival provider. Glasklar, Mullvad and Tillitis cosign the
Sigsum digest with a quorum of two out of three. The qualified timestamp uses
RFC 3161. Its eIDAS Article 41(2) presumption covers the daily checkpoint root
alone. A Verifyum user proof is not a qualified electronic timestamp. Verifyum
is not a qualified trust service. Software Heritage and Internet Archive show
what was stored. They do not establish when the original file existed. The
number of channels is not a quality score.

Every finalized proof is also announced on
[Telegram](https://t.me/verifyum) and in the
[Atom feed](https://verifyum.com/feed.xml). These are announcement channels.
They are excluded from the nine evidence records. Their timestamps date the
announcement. They say nothing about the original file date.

An agent can also assemble one decision record locally from its instructions,
prompt, model, parameters, tool calls and output, then anchor only the
commitment. The proof shows that the exact record existed unchanged by the
block time. It does not prove that the model actually ran with the recorded
settings.

---

## Free public Agent2Agent (A2A) endpoint

Agents that speak Agent2Agent can reach five of these capabilities at:

`https://aisenseapi.com/a2a`

JSON-RPC 2.0, protocol revision 1.0, no account and no API key. The agent card
is a plain GET at `https://aisenseapi.com/.well-known/agent-card.json`.

A2A is the protocol for delegating work to another agent. MCP is the protocol
for exposing tools. Most of this service is tools, so only the five task-shaped
capabilities are offered over A2A: `agent-wake`, `human-approval`,
`agent-inbox` and `webhook-capture`. The other tools are not reachable through
it. MCP stays the richer workflow surface with its own tool schemas. Queue
is available through REST and MCP, not through A2A.

A2A puts no skill id on the wire, so the caller names the skill in a data part
of the message, as `{"skill": "agent-wake", "arguments": { ... }}`. That is a
convention this service documents, not a field the protocol defines, and a
message without it is refused. The card declares `streaming` and
`pushNotifications` false, so those methods answer `-32004` and `-32003`.
`ListTasks` returns an empty page, because nobody is authenticated and a task ID
is the only credential there is. See [`API.md`](API.md#agent2agent-a2a) for the
skills, task states, response shapes and error codes.

---

## SpeedUp (.su) - token cost, measured properly

Every serialization format marketed for LLM input ships a token-saving claim
measured against a single tokenizer. Agent traffic crosses vendors, so we
measured instead of assumed: six formats and 33 single-construct probes,
priced by five vocabulary families (OpenAI o200k, DeepSeek, Qwen,
SentencePiece, Tekken), with each provider's own token counter as ground
truth.

- [`SU-PROFILE.md`](SU-PROFILE.md) - the SpeedUp profile: eight writing
  conventions for agent-to-agent text, each citing its measured price tag
- [`speedup/`](speedup/) - the measurement rig (Python, standard library
  only), corpus, probes and raw numbers; rerun everything with your own keys
- [The study](https://aisense.no/tokenizer-cost-study) - what survived
  measurement, what did not, and why we decided against shipping yet another
  format

Headline numbers: plain TSV averages 0.82x minified JSON across the
five-vocabulary union and beats the token-oriented formats on every model;
declaring columns once and sending values positionally is the entire
mechanism (minus 34 percent per key-value pair); base64 costs 3.9-4.6x the
plaintext it encodes; and single-letter key dictionaries lose to full keys
on four of five vocabularies.

---

## Why this exists

Most utility APIs require sign-up, rate limit tiers, or pricing for basic
operations. This collection skips all of that. Drop a URL into curl, Python,
JavaScript, or an LLM tool definition and it just works.

The collection covers two tiers of usefulness:

- **Workflow endpoints** - the ones that solve real problems in pipelines and agent systems
- **Standard utilities** - hashing, encoding, UUIDs, time, crypto - the building blocks

---

## Three things to know before you write a client

These are service-wide and they decide how your error handling has to look.
This reference combines source-checked contracts with dated production checks.
Queue REST checks and 28-tool MCP discovery were verified in production on 9 September 2026.

**The response key is named after the endpoint.** `/md5_hash` returns
`md5_hash`, `/random_color` returns `random_color`, `/ping` returns `ping`.
There is no generic `data` or `result` wrapper. Do not guess the key -
[`API.md`](API.md) lists every one.

**Errors usually use `{"error": "message"}` with a non-2xx HTTP status.**
Check both the status and the `error` field. The legacy wallet-generation
handlers can return an error object with HTTP 200. Workflow endpoints use
non-2xx statuses, including 409 for conflicts and 410 for an expired record
that has not yet been removed. Once removed, the same ID returns 404. Unknown
routes also return 404. Consult each endpoint for its additional errors.

**There is a rate limit: 5000 requests per IP per 24 hours.** Exceeding it
returns HTTP 429 in the same flat error shape as everything else.

---

## The high-value endpoints

### Heartbeat - know when a worker stops checking in

Create a monitor with an expected check-in interval, a grace period and one
action for a missed deadline. The action can POST to a public webhook or wake
an existing Agent Wake webhook task.

```bash
curl -X POST https://aisenseapi.com/services/v1/heartbeat \
  -H "Content-Type: application/json" \
  -d '{
    "expect_every_seconds": 300,
    "grace_seconds": 60,
    "on_miss": {
      "url": "https://example.com/agent-offline",
      "payload": { "agent": "worker-7" }
    }
  }'
```

The response gives you an unguessable `heartbeat_id`, `ping_url` and
`status_url`. Call the ping URL with POST after each successful cycle.
Each ping moves the expected deadline. It does not move the fixed 24-hour
expiry. A missed deadline fires once, with no retry.

MCP clients can use `create_heartbeat`, `read_heartbeat` and
`ping_heartbeat` for the same state.

Webhook destinations are checked for SSRF at creation and delivery. Private
and reserved addresses, URL credentials, fragments and redirects are blocked.
See [`API.md`](API.md) for the states and response fields.

---

### Lease - one winner for shared agent work

Lease coordinates workers without an account. Mint a private namespace, then
claim a key for a short period:

```bash
curl -X POST https://aisenseapi.com/services/v1/lease/namespace \
  -H "Content-Type: application/json" -d '{}'

curl -X POST https://aisenseapi.com/services/v1/lease \
  -H "Content-Type: application/json" \
  -d '{
    "namespace": "ns_...",
    "key": "invoice:2026-09-05",
    "ttl_seconds": 60,
    "fingerprint": "charge-order-501"
  }'
```

The winner receives an `owner_token` and a monotonic `fencing_token`. A second
worker receives HTTP 409 while the lease is held. The owner can renew, release
or complete the lease with a JSON result. Later callers with the same key and
fingerprint can reuse that completed result.

The lease has a fixed absolute expiry 24 hours after its first acquisition.
Renewals cannot extend it. Raw keys, namespaces, owner tokens and fingerpr