What happened
OhMyLama’s API guide describes its API as OpenAI-compatible and shows a separate web-search endpoint, POST /v1/search. Unlike the chat-completions examples elsewhere in the guide, this endpoint returns ordinary web-search results: links, titles, and excerpts. The documentation explicitly says it does not provide a research mode or generate an answer with a model. A builder can call it with an API key using an HTTP client; the example uses https://ohmylama.ru/v1/search and sends a query and max_results in the JSON body.
The documented query length is 1 to 400 characters. max_results can range from 0 to 20 and defaults to 10. Search depth can be basic, fast, or ultra-fast, with basic as the default. All three modes have the same listed price: 0.01 ruble from the ruble balance per successful request. A successful request is charged even when the result set is empty; failed requests are not charged. The guide says the subscription and API-package tokens are not consumed. The current price can be read from pricing.search_rub through /v1/account.
The endpoint supports topic filters for general, news, and finance; time ranges of day, week, month, or year (also accepted as abbreviated forms); and start and end dates in YYYY-MM-DD format. It can include up to 300 domains or exclude up to 150. An include-domains mode can either restrict or prefer those domains, and requires include_domains. Other listed options include language filtering, publication-date controls, favicons, and exact matching. Setting include_images adds image links; include_image_descriptions adds descriptions when images are included. The guide says image links and descriptions are included in the search price.
There are firm product boundaries. chunks_per_source accepts 1–3 for basic and fast; safe_search: true is available only for basic. Advanced, research, crawl, and extract modes are not offered. auto_parameters, include_answer, and include_raw_content accept only false, and unknown parameters are rejected before search execution. Responses include the query, results with title, URL, content, and score, images, response time, and request ID. Usage reports the ruble cost, remaining balance, and currency. Documented errors include 402 for insufficient funds, 422 for invalid parameters, 429 for rate limits, and 502/504 for search unavailability or timeout without charge.
What the archive adds
New Runtime’s earlier coverage has tracked agent-facing web-search APIs, including DeepSeek’s Search and Fetch APIs (2026-10-22) and Expand Search (2026-10-31). OhMyLama’s documentation is a new provider-specific fact in that broader story: a low-priced endpoint with explicit exclusions and billing rules, rather than a restatement of those launches.
Trend signal
This does not yet fit a pattern New Runtime tracks. The guide exposes a callable search endpoint and controls, but the evidence here is a provider’s API documentation, not a broader shift in how agent protocols or capability boundaries are being adopted.
Why now
The API guide presents web search as an available endpoint and gives builders a way to inspect its current price through /v1/account. It does not explain the timing of the addition, so there is no sourced reason to treat the timing as more than a newly documented capability.
Why it matters
A web-search step can become a costed dependency in an agent workflow rather than an opaque feature bundled into a model response. OhMyLama’s stated per-success price and its distinction between empty results and failed requests give teams a starting point for estimating that dependency and handling retries or no-result cases. Its fixed result cap and explicit absence of research, crawl, and extraction modes also make the intended scope clearer: this is a search-results endpoint, not a complete retrieval or answer-generation system.
That distinction matters in implementation. A team can decide whether it wants to pass titles and excerpts to its own model, apply domain and date restrictions, or use another component for fetching and synthesis. The documentation does not establish that the returned results will meet any particular quality bar, so product teams still need to evaluate relevance and latency on their own queries. The endpoint’s price is useful for comparison, but not a substitute for measuring whether its results work for the task.
Limits and uncertainty
The documented price and capabilities are OhMyLama’s own statements; they have not been independently tested here. The guide does not establish search-result relevance, coverage, latency under load, or availability guarantees. It also does not state a rate-limit threshold, beyond identifying 429 as the error for exceeding one. Builders should verify actual billing, response behavior, and applicable limits in their own accounts before relying on the endpoint in production.
Reader takeaway
OhMyLama documents a narrow, low-cost web-search API: 0.01 ruble per successful request, with a defined set of filters and a hard boundary around what it does not do. Treat it as a candidate source of search results—not as research, crawling, or answer generation—and validate its quality and operational limits against your own workload.