MIAPI
MIAPI is a developer API that answers questions using real-time web search, returning inline citations, sources and a confidence score. It works as a drop-in OpenAI replacement, adds news and image endpoints, and keeps a permanent free tier.
What is MIAPI?
MIAPI is an HTTP API that answers questions with text drawn from a live web search rather than from a model's memory. A single POST to /v1/answer returns a JSON object holding the answer, a list of sources with their titles, URLs and snippets, a confidence value, a cache flag, the query time and a request identifier. Optional parameters shape the result: citations inserts [1][2] markers into the text, response_format switches between plain text, a short form, markdown and JSON, search_domains and exclude_domains restrict or block sites, language forces the language of the reply, and context carries a conversation history for follow-up questions.
The same endpoint carries three modes. Answer synthesises a reply from the web. Search returns raw results with no model processing, which is what a LangChain or homemade RAG pipeline needs. Knowledge answers from up to 10,000 characters of text supplied by the caller, on its own or combined with the web. Around them sit nine further endpoints: an OpenAI-compatible chat completions route, a Server-Sent Events stream that sends the sources before streaming the answer, dedicated news and image searches, a keyless playground, and management routes for models, keys and usage.
Compatibility is the argument the site pushes hardest. An application already written against the OpenAI Python client keeps its code and changes only its base URL, selecting the miapi-grounded model. An official Python SDK, published on PyPI under the MIT licence, wraps the same endpoints in synchronous and asynchronous clients, and a downloadable MCP server plugs the API into assistants such as Cursor, Claude Desktop and Windsurf.
The service is young and thinly documented. Its domain was registered in February 2026, the terms and the privacy policy both date from 28 February 2026, and the whole site runs to four pages. The headline figures come from the vendor alone: around 1.1 seconds per response, 98% accuracy, and $0.0025 per query on the largest pack, alongside a comparison table setting MIAPI against Perplexity, Brave, Exa.ai, Tavily and Google. No underlying language model is named anywhere.
What it does
- Answer a question from a real-time web search in a single POST call
- Return inline [1][2] citations, a source list with titles, URLs and snippets, and a confidence value
- Stand in for the OpenAI chat completions endpoint without touching the application code
- Answer from documents supplied by the caller, up to 10,000 characters, in knowledge mode
- Stream an answer over Server-Sent Events, sending the sources first and then the text token by token
- Return raw search results with no model processing, for LangChain and homemade RAG pipelines
- Search recent news articles and images, with dates, sources, thumbnails and dimensions
When to use MIAPI / When not to
A quick filter to help you decide if MIAPI is the right fit.
When to use MIAPI
- Back-end and full-stack developers who want sourced answers inside their own application without building a search pipeline first
- Teams already calling the OpenAI client, for whom adopting web grounding means changing one base URL
- Builders of RAG pipelines, who can take raw search results or answer from their own documents through knowledge mode
- Newsrooms and monitoring teams, thanks to a dedicated news endpoint returning dates, sources and snippets
- Support and knowledge-base teams answering from their own documentation rather than from the open web
When not to use MIAPI
- Non-technical users: there is no consumer chat interface, everything happens through HTTP calls
- Organisations that require a signed data processing agreement, a formal subprocessor list or a SOC 2 or ISO 27001 attestation, none of which is published
- Projects tied to a specific hosting jurisdiction, since no hosting country or region is declared anywhere
- High-throughput real-time workloads, capped at 60 requests per minute on the free tier and 120 on the paid packs
- Buyers who need an identified legal entity behind the service, and anyone under 16, whom the privacy policy excludes
How to use MIAPI
A typical end-to-end flow, from setup to results.
- Try the playground on the home page: no signup is needed, and the answer format can be set to text, short, markdown or JSON
- Create an account with an email and password, or through Google, then confirm the six-digit code sent by email
- Copy the API key from the dashboard, where an account can hold up to five keys
- Call POST https://api.miapi.uk/v1/answer with an Authorization: Bearer header and a JSON body carrying the question
- Or install the Python SDK with pip install miapi-sdk, then call client.answer() with citations enabled
- Or migrate existing OpenAI code by pointing base_url at https://api.miapi.uk/v1 and selecting the miapi-grounded model
- Pick the mode that fits the job: answer for a synthesised reply, search for raw results, knowledge to answer from your own text
- Tune the call with search_domains, exclude_domains, language, temperature, max_tokens and response_format
- Wire in an MCP assistant: install mcp[cli] and httpx, download miapi_mcp_server.py, set the key and declare the server in Cursor, Claude Desktop or Windsurf
- Track consumption in the Usage tab or through GET /v1/usage, and buy another query pack once the free quota runs out
Pros & Cons
Pros
- Entry pricing among the lowest advertised for grounded answers, at $2.50 to $3.60 per 1,000 queries
- A permanent free tier of 500 queries a month, refilled monthly, with every endpoint open
- Almost no integration work for an existing OpenAI codebase: one base URL to change
- Sources and citations returned systematically, so the end reader can check what is asserted
- Query packs advertised as never expiring, bought once rather than subscribed to
- A public playground with no signup, so the API can be judged before an account exists
- A wide surface for a single API: answers, raw search, news, images, streaming, a Python SDK and an MCP server
Cons
- No identified publisher: no legal entity, no postal address, no legal notice page and no social account
- A very young service, with the domain registered in February 2026 and first archived in May 2026
- No GDPR mention, no data processing agreement, no certification and no declared hosting country
- The site contradicts itself on billing: one-time purchases with no subscription on the home page, monthly Stripe billing in the terms, automatic monthly renewal in the dashboard settings
- Performance figures do not line up either, with roughly 1.1 seconds claimed at the top of the page and 1.5 seconds further down, and a playground limit given as 5 queries per minute on the page but 20 in the endpoint documentation
- No underlying model is ever named, and the accuracy figure and competitor prices come from the vendor alone
- Documentation lives in a single section of the home page, and the dashboard link to the interactive docs returned a 404 error on 11 August 2026
Pricing & Plans
A permanent free plan is available at no cost and covers 500 queries per month with one API key. The cheapest paid entry point is the Starter pack at USD 9.00 for 2,500 queries, that is USD 3.60 per 1,000. All prices are quoted in US dollars.
- 500 queries per month
- 1 API key
- 60 requests per minute
- all endpoints
- refilled monthly
- 2
- 500 queries
- USD 3.60 per 1
- 000
- 5 API keys
- 120 requests per minute
- no expiry
- 10
- 000 queries
- USD 3.30 per 1
- 000
- 5 API keys
- 120 requests per minute
- no expiry
- 30
- 000 queries
- USD 3.00 per 1
- 000
- 5 API keys
- 120 requests per minute
- no expiry
- 100
- 000 queries
- USD 2.50 per 1
- 000
- 5 API keys
- 120 requests per minute
- no expiry
- volume discounts
- unlimited API keys
- custom rate limits
- an SLA guarantee and an on-premise option
- arranged by email with the vendor
Data, GDPR & hosting
A consolidated view of how MIAPI handles your data.
GDPR overview
The GDPR is never named on this site: searching every page collected, in text and in raw HTML, returns no occurrence of it. No legal basis, no data controller, no data protection officer and no Article 27 representative is identified, and no data processing agreement is offered or refused. What the privacy policy does publish are mechanisms that overlap with the regulation without invoking it: a right to request access to, correction of or deletion of personal data, answered within 30 days at support@miapi.uk; usage logs kept for no more than 90 days; a minimum age of 16; and cookies limited to session management and Google Analytics, which the visitor can switch off in the browser. Anyone with a compliance obligation should treat this as an undocumented position rather than a stated one.
Who owns the data?
The terms make the account holder responsible for their API keys and for every call made under their account. MIAPI stores an email address and a hashed password, plus a name when Google sign-in is used; card details stay with Stripe. API queries are logged for usage tracking, and the terms state that they are neither stored long-term nor shared with third parties. Nothing in the terms or the privacy policy claims ownership over the questions submitted, the documents passed in knowledge mode or the answers returned. Users can request access to, correction of or deletion of their personal data at support@miapi.uk, with a stated 30-day response time.
Reuse rights
Nothing in the published documents reserves rights over the generated output, so a developer can reuse answers, sources and citations without asking permission. The limits set out in the terms concern behaviour rather than reuse: the API may serve lawful purposes only, must not be used to produce spam, misinformation or content that breaks applicable law, and must not be reverse-engineered, overloaded or exploited. Two points are left unaddressed. Answers quote third-party web pages whose own rights still apply, and the service is supplied as is, with MIAPI declining responsibility for decisions taken on the strength of its responses and capping its liability at the amount paid over the previous three months.
Data retention & training
Hosting summary
MIAPI publishes no hosting jurisdiction. The privacy policy says only that account data is stored on secured servers, that passwords are hashed, that API keys are generated uniquely per account and that all communications use HTTPS. No country, no region, no cloud provider and no international transfer mechanism is named, and there is no subprocessor page. The closest thing is the third-party section of the privacy policy, which names Stripe for payments, Google for optional sign-in and Google Analytics for traffic measurement. One external observation, which the vendor does not itself claim: on 11 August 2026 the domain resolved to 165.232.45.114, an address geolocated in London, United Kingdom, on DigitalOcean's network. That is where the site answered from on that date, not a commitment about where customer data lives. Anyone with a data residency requirement should obtain it in writing before sending anything.
Things to keep in mind
Risks and trade-offs to weigh before adopting MIAPI.
- Grounding narrows the risk of invented facts without removing it: the No hallucinations promise is a vendor claim, not a measured result
- The confidence value returned with each answer is never explained and should not be read as a guarantee of accuracy
- Cited sources are ordinary web pages of uneven quality, so anything used in journalism, health or legal work still needs a human check
- Leaning on a synthesised answer without ever opening its sources slowly erodes the habit of verifying anything at all
- The publisher is anonymous and the domain expires in February 2027, which makes a production dependency a continuity risk
- No underlying model is named, so answers can shift without notice if the vendor changes it
- With no data processing agreement and no declared hosting country, sending personal or confidential text through knowledge mode exposes the sender, while liability is capped at the last three months of spend and no availability is guaranteed
Setup & Integrations
Technical difficulty
Straightforward for a developer, out of reach without code. Testing costs nothing, since the home page playground needs no account. Going live means creating an account, copying a key and sending one POST request with an authorisation header; the Python SDK installs in a single pip command, and an existing OpenAI integration only needs its base URL changed. The MCP server takes a few more manual steps: install two packages, download a Python file, set the key and declare the server in the assistant. There is no no-code interface of any kind.
Deployment
Integrations
Supported languages
Behind MIAPI
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement MIAPI.
Frequently asked questions
What does a MIAPI call actually return?
Do I have to rewrite my existing OpenAI code?
Can I test it without creating an account?
How much does it cost?
Do purchased queries expire?
Can it answer from my own documents?
What are the rate limits?
What happens to my data?
Is the service GDPR compliant?
Is there an SDK?
Should you pick MIAPI?
MIAPI does one thing and states it plainly: it turns a question into an answer backed by a live web search, with the sources attached. For a developer, the appeal is the shortest possible path to that result, one POST call or a single base URL change in an application already written for OpenAI, at an advertised $2.50 to $3.60 per 1,000 queries and with 500 free queries a month to judge it on. The surface is unusually wide for the price: synthesised answers, raw search for RAG pipelines, answers drawn from your own documents, streaming, news and image endpoints, a Python SDK and an MCP server.
The reservations concern the vendor rather than the product. Nobody is named behind it: no company, no address, no legal notice, no social account, and only two email addresses for contact. The domain was registered in February 2026 and expires in February 2027. Nothing is published about hosting jurisdiction, data processing agreements or certifications, and the GDPR is never mentioned. The site also contradicts itself where it matters most, promising one-time purchases with no subscription on the home page while the terms describe monthly billing through Stripe. The performance claims are the vendor's own, and no underlying model is named.
The sensible reading is that MIAPI is a cheap, well-shaped building block worth testing rather than an infrastructure supplier to commit to blindly. Start on the free tier, measure latency and source quality against your own questions, and keep a fallback in the integration. Teams under contractual compliance obligations should wait until the publisher identifies itself and documents where the data goes.
- Choosing a selection results in a full page refresh.
- Opens in a new window.