API docs
Send up to 100 texts at a time and get one answer per text: yes or no, the best label from your list, or the team a support ticket belongs to. Each answer comes with a score, so you choose how sure it must be.
Get access in seconds. Every account comes with free texts to try it.
One yes/no question about each text, read by the smart model. $3 per 10,000 texts.
The best label from your list of 2 to 200. $1 per 10,000 texts.
The team for each support ticket; unsure ones go to a person. $1 per 1,000.
per new text on a plain CPU. Repeat texts come back from cache.
Every tool is one POST to https://fastdar.com/mcp with your key in the Authorization header. The body is a JSON-RPC tools/call, the same format AI assistants use, so one integration serves your code and your assistant. The answer is in result.structuredContent.
curl https://fastdar.com/mcp \
-H "Authorization: Bearer $FASTDAR_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "tools/call",
"params": {"name": "classify", "arguments": {
"labels": ["billing", "bug", "how-to", "sales"],
"texts": ["Can I pay yearly instead of monthly?"]}}}'
{"jsonrpc": "2.0", "id": 1, "result": {"isError": false,
"structuredContent": {"results": [{"index": 0, "label": "billing", "probability": 0.8992}]},
"content": [{"type": "text", "text": "..."}]}}
Below, only the arguments and the structuredContent are shown. "method": "tools/list" returns every tool with its full input schema.
{"question": "Is this review a complaint?",
"texts": ["Arrived broken, second time this month.", "Love it, works great."],
"yes_examples": ["Stopped working after two days.", "Support never replied to me.", "Wrong size and they won't take it back."],
"no_examples": ["Great quality, fast delivery.", "Exactly what I wanted.", "Five stars, would buy again."]}
{"results": [{"index": 0, "answer": "yes", "p_yes": 0.9424},
{"index": 1, "answer": "no", "p_yes": 0.0211}],
"cutoff": 0.9, "tier": "fast"}
On fastdar.com, check is answered by the smart model, which reads the question itself: up to 25 texts per call, $3 per 10,000, and examples are optional. Where only the fast model runs, give three or more examples of each answer, because without them it is unsure on most questions. tier says which one answered; the output above is from the fast model.
{"labels": ["billing", "bug", "how-to", "sales"],
"texts": ["Can I pay yearly instead of monthly?", "The export button does nothing."],
"min_confidence": 0.6}
{"results": [{"index": 0, "label": "billing", "probability": 0.8992},
{"index": 1, "label": "not_sure", "probability": 0.5637, "best_guess": "how-to"}]}
A label can be a string or {"label": ..., "description": ..., "examples": [...]}. Below min_confidence (default 0) you get not_sure and the best_guess.
{"topics": [{"label": "Payments", "examples": ["My card was declined", "I was charged twice"]},
{"label": "Shipping", "examples": ["Where is my parcel?", "Tracking hasn't moved"]},
{"label": "Accounts", "examples": ["I can't log in", "Reset my password"]}],
"tickets": ["Card declined at checkout", "Package stuck in transit for a week"]}
{"results": [{"index": 0, "route": "Payments", "topic": "Payments", "confidence": 0.9619},
{"index": 1, "route": "Shipping", "topic": "Shipping", "confidence": 0.8437}],
"to_person": 0}
Tickets below min_confidence (default 0.7) get "route": "person", with the topic it would have picked. Up to 500 topics; more past tickets per topic make it sharper.
Save your labels once, use them by name, and tell us when an answer was wrong. Each correction becomes an example for the right label, so the set gets more accurate on your own text. Corrections are free.
Give each label a name and a few real examples, like past messages that belong there.
Send messages, tickets or reviews, and each one comes back with the best label and how sure it is.
Tell FastDAR the right label. That text becomes a new example, so similar texts are labelled right next time. You can turn learning off for a set when you're happy with it.
{"name": "support",
"labels": [{"label": "Payments", "examples": ["My card was declined"]},
{"label": "Shipping", "examples": ["Where is my parcel?"]},
{"label": "Accounts", "examples": ["I can't log in"]}]}
Then pass "label_set": "support" to classify or sort_ticket instead of listing the labels. Saving again under the same name replaces the set.
{"label_set": "support", "text": "Refund hasn't arrived yet", "label": "Payments"}
{"name": "support", "label": "Payments", "examples": 2, "corrections": 1}
No code? Make a set on your account page and label your Gmail with it: a short script in your Google account labels new emails, and moving an email to the right label teaches the set. Using Zendesk, Freshdesk or Help Scout? Tag new tickets with your set.
Learning is on for a new set. Turn it off when you're happy with how a set answers: it then keeps answering exactly as saved, and corrections are refused until you turn learning back on. Use the switch on your account page, or the API:
PUT /v1/labels/support/learning {"on": false}
Your saved sets stay with your account when you make a new key, and your account page shows how many corrections each label has learned from. They're deleted when you delete your account. Up to 50 sets, 200 labels a set and 200 examples a label; the oldest examples go first.
| Assistant | Setup |
|---|---|
| Claude Code | /plugin marketplace add FastDAR/claude-plugin then /plugin install fastdar@fastdar |
| ChatGPT | Settings › Apps & Connectors › Advanced › Developer mode, then Create. Address https://fastdar.com/mcp, sign-in OAuth. |
| claude.ai | Settings › Connectors › Add custom connector, address https://fastdar.com/mcp. |
| Cursor | Cursor Settings › MCP › Add new MCP server, or in ~/.cursor/mcp.json: {"mcpServers": {"fastdar": {"url": "https://fastdar.com/mcp", "headers": {"Authorization": "Bearer YOUR_KEY"}}}} |
| Gemini CLI | gemini extensions install https://github.com/FastDAR/claude-plugin |
ChatGPT and claude.ai sign in with your FastDAR email and ask which key to use, so no key is pasted into a chat. You can disconnect an app any time from your account page.
| What | Limit |
|---|---|
| Texts per call | 100 (25 when the smart model answers check), each up to 4,000 characters |
| Free allowance | 10,000 texts for sorting and labelling and 200 texts for yes/no checks, once per email |
| Out of credits | The result has "isError": true and says how to add credits (HTTP 402 on the other endpoints). We email you before it runs out. Top up and the same key carries on. |
| Bad input | The result has "isError": true and says what to fix |
| Endpoint | What it does |
|---|---|
GET /v1/usage | Your plan, credit balance and what each kind of call has cost. |
POST /billing/topup | Add credits: {"amount_usd": 25} returns a Stripe checkout link. |
POST /v1/decide | Several questions about one text in a single call: {"text": "...", "questions": {"urgent": {"type": "yes_no", "question": "Is this urgent?"}}}. Question types are yes_no, choice and score, with optional hand-off to the reasoning model. |
Pay as you go with prepaid credits from $10, after the free texts run out. See all prices.
Answers come from small models and can be wrong. Measure them on your own texts before you rely on them, and keep a person in the loop for decisions that matter.
We store your email, a hash of your key and usage counts. We do not store the texts you send.
© FastDAR · About · Support · Terms · Privacy · Data processing
FastDAR is a service of Media Yard LLC, PO Box 73, Pennsauken, NJ 08110, United States.