SaaS, AI & Developer Infrastructure
Web Research API
Your agents need facts from outside your product. The hard part is not fetching them — it is returning something your customers can audit, and knowing which sources you are entitled to use.
- AI agent platforms
- Assistant and copilot vendors
- Research tools
- Sales intelligence software
- Developer platforms
What your customers would see
example interfaceMockupIllustrative mockup of the interface we would build. Figures shown are examples, not results from a live deployment.
The problem this solves
Every agent platform reaches the same point. The model is fine, the orchestration works, and then a customer asks a question whose answer lives on the open web. The usual fix is to wire in a search API, concatenate a few snippets into the context window and hope. It demos well. It fails in production because nobody can tell where an answer came from, how old it was, or whether two sources contradicted each other.
A research layer worth embedding returns structure rather than prose: claims with the source behind each one, a retrieval timestamp, an indication of whether sources agreed, and an explicit statement when the question could not be answered from what was found. Agents can act on that. They cannot act on a paragraph of plausible text.
There is also a sourcing question that agent platforms consistently underestimate. Search APIs, licensed news archives, open datasets and directly fetched pages all carry different terms about caching, redistribution and model training. A platform that passes third-party content through to its customers as part of a paid product is redistributing it, whatever the architecture diagram says. Working out which sources permit that is discovery work, and it is better done before the contract than after a takedown notice.
Who buys this
AI agent platforms, Assistant and copilot vendors, Research tools, Sales intelligence software, Developer platforms
What the API returns
What comes back
- Research answers decomposed into individual claims, each with its sources
- Source references with publisher, URL, publication date and retrieval timestamp
- Freshness metadata per claim, so an agent can tell a 2019 figure from last week's
- Agreement and contradiction signals where sources disagree, rather than one picked silently
- Source tier labelling: primary publication, official register, reputable secondary, user-generated
- An explicit insufficient-evidence response when the question cannot be answered from permitted sources
- Per-source usage rights in the response, so your platform knows what it may store and show
- Retrieval traces for audit — what was queried, fetched, kept and discarded
Endpoints
| Method | Path | Returns |
|---|---|---|
| POST | /v1/research | Run a research query and return structured claims with sources |
| POST | /v1/research/async | Submit a deeper multi-step research job with a webhook callback |
| GET | /v1/research/{id}/trace | Full retrieval trace for audit and debugging |
| POST | /v1/verify | Check a specific claim against permitted sources |
| GET | /v1/sources | Configured sources with licence terms and permitted uses |
| POST | /v1/freshness | Re-check whether previously returned claims still hold |
Request and response
illustrativeRequest
{
"tenant_id": "agentcloud",
"query": "What is the current statutory corporate tax rate in Ireland and when did it last change?",
"depth": "standard",
"max_sources": 6,
"require_tiers": ["primary", "official_register"],
"answer_mode": "claims"
}Response
{
"research_id": "res_9a2201",
"retrieved_at": "2026-10-22T09:41:12Z",
"elapsed_ms": 6840,
"status": "answered",
"claims": [
{
"claim": "Ireland applies a 15% minimum effective rate to in-scope large groups.",
"confidence": 0.93,
"agreement": { "supporting": 3, "contradicting": 0 },
"sources": [
{ "publisher": "Irish Revenue", "tier": "primary",
"published": "2026-01-08", "retrieved": "2026-10-22T09:41:05Z",
"usage": { "quote_permitted": true, "cache_days": 30, "redistribute": false } }
]
},
{
"claim": "The 12.5% headline rate remains for groups below the EUR 750m threshold.",
"confidence": 0.88,
"agreement": { "supporting": 2, "contradicting": 1 },
"sources": [
{ "publisher": "Department of Finance", "tier": "primary",
"published": "2025-10-01", "retrieved": "2026-10-22T09:41:07Z",
"usage": { "quote_permitted": true, "cache_days": 30, "redistribute": false } }
],
"note": "One secondary source states 12.5% without the threshold qualification."
}
],
"insufficient_evidence_for": [],
"sources_excluded": [ { "reason": "terms prohibit automated access", "count": 2 } ]
}Where the data comes from
| Licensed search APIs | Commercial search providers whose terms set out what may be cached, quoted and shown to end users — the terms are the product, not an afterthought |
| Official registers and primary publications | Regulators, statistical agencies, company registries and government portals, which are both authoritative and generally the clearest on permitted reuse |
| Open datasets and licensed archives | Open-licence corpora and commercially licensed news or journal archives, where the licence names the permitted use |
| Directly fetched pages within policy | Only hosts whose robots directives and terms permit automated reading, with rate limits and caching that respect them |
| Your customers' own content | Where a tenant connects its own documents, that content is theirs and carries no third-party licensing problem |
| Source quality and tier metadata | Publisher identity and reliability labelling, maintained deliberately rather than inferred from domain authority alone |
The intelligence layer
The difference between a demo and a service is mostly in this section. These are the parts that keep the answer trustworthy once real traffic is hitting it.
- Query decomposition into sub-questions, since a single search rarely answers a compound question well
- Claim-level extraction with source attribution, so the unit of output is a verifiable statement rather than a paragraph
- Cross-source agreement checking, with contradictions reported rather than resolved by whichever source ranked first
- Freshness handling that distinguishes publication date from retrieval date and flags figures whose currency matters
- A per-source policy layer enforced in code: robots directives, rate limits, cache windows and redistribution rights evaluated before a fetch and again before a response
- Grounding validation — every returned claim must be supported by retrieved text, and unsupported claims are dropped rather than softened
- Accuracy measured against a held-out question set with known answers, reported per source tier and per question type
- Monitoring for source drift, blocked hosts, licence expiry and rising insufficient-evidence rates
What it changes for the business
Auditable agent output
Claims with sources and timestamps give your customers something to check, which is usually what blocks an agent pilot from going into production.
Fewer confident errors
Contradiction reporting and explicit insufficient-evidence responses remove the failure mode that damages trust fastest.
Usage-based pricing that fits
Research volume scales with your customers' agent activity, so revenue grows with their adoption rather than with your sales effort.
A sourcing position you can defend
Knowing which sources you may redistribute is a question enterprise procurement will ask, and a clear answer shortens the review.
How it fits into your product
How it embeds
- White-label API behind your own agent runtime, with no MLPivot attribution in responses
- Synchronous endpoint for interactive queries, asynchronous jobs with webhooks for deeper research
- Per-tenant keys, quotas and cost caps, so a runaway agent loop cannot generate an unbounded bill
- Tool-call schemas and function definitions ready to register with your existing agent framework
- Retrieval traces exposed to your customers where they need to audit what their agents read
- Region-pinned deployment and configurable caching where data residency or licence terms require it
Included in every build
- Authentication, scoped API keys and per-customer rate limiting
- Input validation with predictable, documented error responses
- Versioned endpoints so your integration never breaks without notice
- Request and response logging for audit and dispute resolution
- Latency, error-rate and usage monitoring with alerting
- OpenAPI specification, sandbox environment and client examples
- Backups, failover design and a documented recovery procedure
- An agreed support arrangement once the service is live
How we would build it
about 20 weeksDiscovery and feasibility
Weeks 1–3
This one depends heavily on data access, so we start by proving the sources exist, can be licensed for your use, and actually carry the signal the model needs. If they do not, you find out here.
Architecture and baseline model
Weeks 4–6
A written architecture and a baseline model measured against a holdout period, so you know what accuracy you are buying before the full build starts.
Production build
Weeks 7–16
Data pipelines, model training and serving, historical archive, validation harness, API, security, monitoring and documentation.
Validation and integration
Weeks 17–20
Backtesting against held-out history, accuracy reporting your team can audit, integration support and an agreed retraining cadence.
Commercial models
There is no fixed price list, because the cost is driven by data complexity, integration count, model requirements, request volume and the support level you need. Scope is agreed in writing before development starts.
Fixed-scope prototype
A defined slice of the service, built and priced as one piece of work. The usual first step when the business case needs proving internally.
Fixed-price build
An agreed scope against a written architecture, delivered for an agreed price. Best when the requirements are already clear.
Build then retainer
We build the service, then operate and improve it under a monthly agreement covering monitoring, updates and model maintenance.
White-label product
The service runs under your brand, your domain and your pricing, with MLPivot invisible to your customers.
Questions buyers ask
Will you fetch any site our agents ask for?
No. Hosts that disallow automated access in robots.txt or prohibit it in their terms are not fetched, bot protections are not circumvented, and per-host rate limits apply regardless of what an agent requests. The response names which sources were excluded and why, so your customers see a policy decision rather than an unexplained gap.
Can our customers store and display the content you return?
That depends entirely on the source, which is why usage rights are returned per source rather than stated once in our documentation. Many licensed search and news sources permit short quotations with attribution and a cache window of days, and prohibit redistribution or model training. The API returns those constraints as data so your platform can enforce them; building a product that ignores them is how agent platforms acquire legal problems.
How is this different from calling a search API ourselves?
Mostly in what comes back. A search API returns ranked links and snippets; this returns claims with attribution, freshness, agreement signals and an auditable trace, plus the policy layer that decides what may be fetched and reused. If your customers are happy with links, you do not need this. If they are asking where an answer came from, you do.
Does it ever just say it does not know?
Yes, and that behaviour is a requirement rather than a fallback. A question that permitted sources do not answer returns insufficient evidence for the specific sub-question, not a hedged paragraph. Tuning that threshold against your own question set is part of the build.
How accurate is it?
Accuracy depends far more on the question type than on the model. Questions with a single authoritative publisher — statutory rates, filing dates, published standards — resolve reliably. Questions requiring synthesis across contested or sparse sources do not, and we would measure both on your own evaluation set during the prototype rather than quoting a headline number.
Related blueprints
Enterprise Search API
Search across a customer's connected documents and structured records, built as the search layer inside your own product.
Website Change Monitoring API
Alerts when pages your customers are permitted to monitor change — prices, terms, policies, stock or staff listings — as a feature inside your product.
Document Extraction API
Structured fields pulled from PDFs, scans and forms, built as the extraction layer your ERP, CRM or accounting product embeds.
Want this built for your platform?
Tell us what your product does today and what you want another system to be able to ask it. We will come back with an approach, a scope and a price.