Glasser has launched a pay-per-call data service for AI agents, putting around 1,500 paid API endpoints from 29 data providers behind one key and one prepaid balance. Instead of subscribing separately to SEO tools, contact databases and search APIs, an agent can find the right endpoint, check the exact price, run the call and pay only for that call, often a fraction of a cent.
The service drew attention on 29 September when TestingCatalog posted a hands-on test: two real jobs, an Apollo company lookup and a DataForSEO keyword report, cost a combined $0.038 out of the $1 of free credit new accounts receive. Launch-day posts from early users put a Google search at $0.0011 and an Ahrefs keyword report at $0.66.
This article explains what Glasser launched, how its pricing and charging rules work, what the catalogue really contains, what common calls cost, when it beats buying data direct, what early testers found, the terms that matter, and who should try it.
Table of contents
What Glasser Launched
Glasser describes itself as “pay-as-you-go APIs for AI agents”. The company behind it is Super Intent, Inc., according to the site’s footer and terms. Its documentation is blunt about the model: “Glasser is a broker. You buy individual API calls from third-party Providers through one Key and one prepaid Balance. We do not make the data; Providers do, under their own terms.”
One key and one balance
The pitch is aimed at a real annoyance for anyone running AI agents. Premium business data, such as keyword rankings, company records or verified work emails, usually sits behind monthly plans or minimum deposits. An agent that needs ten lookups from five different providers would otherwise need five accounts, five API keys and five bills. The new service replaces that with one key, one workspace balance and a price shown before each call.
Setting up an agent
Setup is a single instruction. Users paste “set up https://glasser.ai/SKILL.md” into Claude Code, Codex, Cursor or another supported agent. According to TestingCatalog, the agent then installs the command-line tool and opens a browser window to sign in. From there it browses the catalogue itself, picks an endpoint for the job, checks the cost, runs it and reports what it spent.
The same service is available four ways: the agent skill, a command-line tool, a remote MCP server with seven tools (search, inspect, run, three run-management tools and balance), and a plain HTTP API. The documentation stresses that all four are “one implementation behind four doors”, sharing the same balance and spending policy.
Why the company built it
A blog post from 12 September sets out the thinking. “The old rule still applies: garbage in, garbage out,” it says, arguing that AI makes bad data harder to spot because it turns incomplete information into answers that sound convincing. The company wants to become “one Data Router” that picks the right provider, falls back to another or combines sources. The name, it says, means “giving AI a pair of glasses”.
The site does not name the founders or disclose funding, which is worth noting for a service that will hold prepaid balances.
How Pay-Per-Call Pricing Works
The pricing model is simple on the surface and careful underneath. It is also better documented than many API marketplaces, which helps when an agent, rather than a person, is spending the money.
Search, inspect, run
Every call follows the same three steps. The agent searches the data sources, inspects an endpoint to see its input format and exact price, then runs it. Prices are either flat per call or per result with a cap. Glasser’s documentation says money is always handled as an exact decimal string, never as a floating-point number, so rounding errors cannot creep into a bill.
Before anything is sent to a provider, the service checks the workspace’s policy, quotes the price and places a hold on the balance. If the request is refused because of bad input, low balance or policy, no run is created and nothing is charged.
What you pay when a call fails
The most useful design choice is that every price comes with a charge clause for each outcome. The documentation insists that a run’s status and its charge are “independent facts”, so an agent must read both.
| Outcome | What it means | How it is charged |
|---|---|---|
| Refused before dispatch | Bad input, insufficient balance or policy block | No run is created; nothing is charged |
| Completed with data | The provider answered with a result | The published price |
| Completed with no result | The provider answered “not found” | The endpoint’s no-result clause |
| Failed | Provider error, timeout or internal error | Per the matching clause, which can be non-zero |
| Stopped | Cancelled before dispatch | Not charged; a run already sent completes and is charged |
Retries and runaway agents
Agents retry things, especially after timeouts, and a naive retry can buy the same data twice. Every run here carries an idempotency key chosen by the caller. Sending the same key again returns the original run, marked “replayed”, and nothing new is charged. That is a small detail with real money behind it when an agent loops.
Spending controls sit at the workspace level. Each key belongs to a workspace, and the effective policy is the overlap of the workspace’s rules and the key’s own. An endpoint outside a key’s policy simply does not exist for that key: search never lists it and inspect answers “not found”.
What Is Actually in the Glasser Catalogue
Launch coverage and the company’s own posts describe “~2000” endpoints. The public site map, last updated on 23 September when we checked on 30 September, lists 1,558 endpoint pages from 29 providers, the same figures one early user quoted on launch day. The gap may be endpoints added since, or entries not yet published as pages.
The mix is uneven. Nearly half of the listed endpoints come from a single provider, TikHub, which covers Douyin, TikTok, Xiaohongshu and other social platforms, with one request per query.
The names business users will look for
The big business-data brands are present but thin. Apollo has 9 listed endpoints, Semrush 4, ZoomInfo 4 and People Data Labs 12, against 38 for Ahrefs and 207 for DataForSEO. Other providers include Bright Data, Apify, BuiltWith, Crustdata, Exa, Hunter, Keepa, LeadMagic, Lusha, OpenAlex, PredictLeads, Prospeo, RentCast for US property, SEC API for filings, Serper, Serpstat, TheirStack and TypeSafe’s Jev decision model.
So the headline count flatters the catalogue for most business uses. For SEO, lead research and company intelligence, the useful set is a few hundred endpoints, not two thousand. That is still a lot of data behind one key.
What Common Calls Cost
We checked a sample of endpoint pages on 30 September. Prices are the service’s published per-call or per-result rates and can change between runs; the terms say the price that applies is the one shown when a run is requested.
| Provider and endpoint | Listed price |
|---|---|
| ZoomInfo company search by industry, revenue or location | $0.0005 per call |
| Serper Google web search | $0.0011 per call |
| Semrush authority score and backlink totals | $0.002475 per call |
| Exa answer from web sources with citations | $0.005 per call |
| DataForSEO Labs ranked keywords | $0.01548 per call |
| Apollo company news or open job postings | $0.026 per call |
| TheirStack job postings search | $0.035934 per result |
| People Data Labs person enrichment | $0.30 per call |
| ZoomInfo company enrichment | $0.495 per result |
| Ahrefs organic keywords | $0.66 per call |
Some calls are free. A Hunter endpoint that counts email addresses at a domain by department and seniority was listed at $0.00 per call, which lets an agent check whether a paid lookup is worth making.
Is Glasser Cheaper Than Going Direct?
The honest answer is that it depends on volume. The homepage makes its own comparison, “checked September 18, 2026”: DataForSEO requires a $50 minimum deposit for paid use, Serper’s Starter plan gives 50,000 credits for $50, and Ahrefs Lite includes API access limited to 100 rows per request.
Search: a small premium, no commitment
Serper’s own pricing page lists Starter at $50 for 50,000 queries, or $1.00 per thousand, with credits valid for six months. Glasser charges $0.0011 per search, or $1.10 per thousand, about 10% more per query. In exchange there is no $50 upfront. At $0.0011, $50 buys about 45,450 searches, so anyone using fewer than that in six months pays less through the broker.
At high volume the gap widens. Serper’s largest listed plan works out at $0.30 per thousand queries, so a heavy user would pay about 3.7 times more per search through the broker.
SEO data: the Ahrefs maths
Ahrefs lists its Lite plan at $129 a month. At $0.66 per organic-keywords call, an agent could make about 195 of those calls a month before Lite became cheaper on price alone. Lite also includes the full Ahrefs web app for a person to use, which the broker does not, so the comparison only holds for teams that genuinely want occasional machine access.
| Data source | Buying direct | Through the broker | Break-even |
|---|---|---|---|
| Serper search | $50 for 50,000 queries, valid 6 months | $0.0011 per query | About 45,450 queries per 6 months |
| Ahrefs | Lite plan $129 a month | $0.66 per organic-keywords call | About 195 calls a month |
| DataForSEO | $50 minimum deposit | From fractions of a cent per task | Depends on the mix of tasks |
The bigger saving is not per call but in the subscriptions you never start. A team that needs SEO data one week, company data the next and funding data the week after would otherwise pay three monthly minimums for bursts of use.
What Early Testers Found
TestingCatalog’s test gives the most concrete independent numbers. An Apollo lookup on cursor.com returned in under half a second with headcount, funding rounds, investors, a department breakdown and 132 detected technologies, for $0.026. A DataForSEO search on TestingCatalog’s own domain found 2,995 keywords it ranks for on Google in the US and returned the top 100 with positions and search volume, for $0.012.
A launch-day thread by X user Prajwal Tomar was more enthusiastic in tone. He reported that researching three EU fintechs cost 30 cents and one full company record 2.6 cents. The company’s own demos claim that finding 48 freshly funded AI agent start-ups and shortlisting 10 cost $1.06, and that more than 500 sales leads with 287 emails cost $7.03, or about $0.0245 per email found.
What the tests do not show
These are small jobs chosen to show the service at its best, and none measures data quality. The broker’s own terms say it “does not create, own, verify, or control the data returned by Providers”, and that outputs “may be incorrect, outdated, or incomplete”. Cheap access to a provider’s data is only as good as that provider.
The Small Print: Terms, Privacy and Data Rules
For businesses, the terms matter as much as the prices. Several points stand out.
| Topic | What the terms say |
|---|---|
| Refunds | Top-ups are non-refundable except for billing errors, service shutdown or account closure without cause |
| Expiry | Paid balance does not expire; promotional credits do |
| Your agents | You are responsible for every agent you authorise “as if the Agent’s actions were your own” |
| Personal data | You are the controller; the broker is your processor; GDPR, UK GDPR, CCPA and anti-spam laws apply to you |
| Reuse | No reselling raw outputs or building a competing database from them |
| Training and sale | Run data is not used to train models and is not sold |
| Disputes | Binding arbitration with a class-action and jury-trial waiver |
Agents you did not mean to run
The agent clause deserves attention. The terms make users responsible for configuring agents to inspect prices, respect spending limits and avoid “speculative, duplicative, or bulk Runs”, and say the company “is not responsible for Runs requested by an Agent that you did not intend”. Workspace policies and idempotency keys help, but the financial risk of a looping agent sits with the customer.
People data and UK GDPR
Many endpoints return personal data: names, employers, emails and phone numbers. The privacy policy, updated on 2 September, says providers do not see which customer made a request, but that providers “may retain query logs” under their own terms. For UK firms using lead data, the lawful basis, transparency to the people concerned and direct-marketing rules remain your job, exactly as if you had bought the data yourself. Cybersecurity and compliance teams should review which endpoints agents are allowed to call before a key is issued.
One small inconsistency is worth flagging. The documentation says a free grant “does not expire”, while the terms say promotional credits do. Anyone relying on referral or share grants should check which applies.
How Glasser Fits the Agent Data Market
The launch is part of a wider shift towards services built for AI agents rather than people. Other projects are experimenting with agents paying per request in stablecoins, while this broker keeps a conventional card-funded balance through Stripe. Its bet is that businesses want agent access without adopting a new payment rail.
It is also spreading through agent platforms quickly. It is listed in the Hermes Agent plugin catalogue and on ClawHub, and the company says its data is available in the Dify Marketplace, and it has published 13 open-source skills on GitHub for jobs such as competitor research and SEO audits. On the day OpenAI launched Dots at DevDay, its X account posted: “Looks like Dot and Glasser already see eye to eye.” We have covered similar tools that put specialist data inside agents, including SpyTrend’s ad-intelligence MCP server and a Google Maps scraper MCP for AI agents.
The router is still a promise
The “Data Router” idea, automatically choosing and combining providers based on price and quality, is the most ambitious part of the pitch. Today the agent does the choosing, guided by prices and descriptions. If the company can add reliable quality scores per provider, that would be worth more than any single price cut.
Who Should Try It
The service suits small teams, consultants and developers who need occasional access to several premium data sources through agents, and who would otherwise pay monthly for tools they barely use. The free $1 credit, with no card needed, is enough to test real jobs.
It suits heavy users of a single source less well. If your agents run tens of thousands of searches a month, or live inside Ahrefs all day, buying direct will usually be cheaper. Larger organisations will want clarity on who runs Super Intent, service levels and data processing terms before routing sensitive lookups through a new broker. For help deciding where agents fit in your own processes, see our work on AI employees and autonomous AI agents.
Glasser FAQ
What is Glasser?
Glasser is a pay-per-call data broker for AI agents, run by Super Intent, Inc. It resells individual calls to around 1,500 paid endpoints from 29 providers, including Ahrefs, Apollo, DataForSEO, Serper and ZoomInfo, through one key and one prepaid balance.
How much does it cost?
There is no subscription. Each endpoint has a published price, from free or fractions of a cent up to $0.66 for an Ahrefs organic-keywords call in our sample. New workspaces get $1 of free credit without a card.
Which agents can use it?
Any agent that can follow its setup file, including Claude Code, Codex and Cursor, plus clients of its MCP server and anything that can call its HTTP API or command-line tool.
Do failed calls cost money?
It depends on the endpoint’s charge clause. Refused requests are free, and retries with the same idempotency key are not charged twice, but a provider “not found” answer or some failures can carry a charge.
Is it cheaper than buying the data directly?
For occasional use across several providers, usually yes, because there are no monthly minimums. For heavy use of one provider, buying direct is usually cheaper per call.
References
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
