Skip to main content
Glama

срезAI — Search API for AI agents

Глубокое исследование / Deep research

deep_research
Read-only

Проводит многошаговое исследование: сам формулирует запросы, ищет, читает источники и возвращает готовый связный ответ со ссылками на использованные страницы.

Когда: вопрос требует сопоставления нескольких источников и вывода — «сравни», «разберись», «что известно о». Когда не: нужен один факт или список ссылок — это web_search, он в 20 раз дешевле и отвечает за секунды. Как это работает: исследование идёт 10–40 секунд. Инструмент сам запускает конвейер — переформулировка вопроса, параллельный поиск, семантическое ранжирование, чтение источников и синтез — и возвращает готовый ответ в том же вызове. Ждать и дозабирать результат по талону больше не нужно. Если набор источников оказался неполным (сработал предел по времени), в ответе будет честная пометка об этом — такой вывод стоит перепроверить через verify_claim. Запасной путь: если конвейер недоступен, инструмент переходит на прежний воркфлоу (3–40 минут) и возвращает строку с пометкой [research_pending] и талоном (ticket) — тогда вызовите инструмент ещё раз с этим ticket (query можно не повторять), чтобы забрать результат. Дозабор по талону НЕ списывается заново — деньги берутся один раз, за саму задачу. Возвращает: текст ответа плюс список источников. Если источники не вернулись, в ответе будет предупреждение — такой вывод не считается проверенным. Цена: 20 кредитов плюс 3 за каждую 1000 токенов ответа — самый дорогой вызов. Списывается один раз за задачу; повтор ТЕМЫ (новый query) — новая задача и новое списание, а дозабор по ticket бесплатен.

Runs multi-step research: forms its own queries, searches, reads sources and returns a finished answer with links to the pages it used.

Use when: the question needs several sources reconciled into a conclusion — "compare", "analyse", "what is known about". Do not use when: you need a single fact or a list of links — that is web_search, 20× cheaper and seconds fast. How it works: research takes 10–40 seconds. The tool runs the pipeline itself — query rewriting, parallel search, semantic reranking, source reading, synthesis — and returns the finished answer in the same call. No hour-long waits and no ticket polling. If the source set turned out incomplete (the wall-clock limit hit), the response says so honestly — verify such output with verify_claim. Fallback: if the pipeline is unavailable the tool switches to the previous workflow (3–40 minutes) and returns a string marked [research_pending] with a ticket — call the tool again with that ticket (you may omit query) to collect the result. Polling by ticket is NOT charged again — you pay once, for the job itself. Returns: the answer text plus a source list. If no sources came back the response says so — treat that output as unverified. Cost: 20 credits plus 3 per 1000 output tokens — the most expensive call. Charged once per job; repeating the TOPIC (a new query) is a new job and a new charge, while polling by ticket is free.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
queryNoВопрос или тема исследования. Обязателен для НОВОГО исследования; при дозаборе по ticket не нужен. / The question or research topic. Required to START research; omit it when polling by ticket.
ticketNoТалон незавершённой задачи из прошлого ответа `[research_pending]`. Передайте его, чтобы дождаться готового результата — бесплатно, без нового списания. / Ticket of an unfinished job from a previous `[research_pending]` response. Pass it to wait for the finished result — free, no new charge.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changed
    • changedInput schema / properties / ticket / type
      Previous value: -"string"New value: +[
      +  "string",
      +  "number"
      +]
  2. Changed4 schema fields changed
    • changedInput schema / properties / query / description
      Previous value: -"Вопрос или тема исследования / The question or research topic"New value: +"Вопрос или тема исследования. Обязателен для НОВОГО исследования; при дозаборе по ticket не нужен. / The question or research topic. Required to START research; omit it when polling by ticket."
    • removedInput schema / properties / query / minLength
      Removed value: -1
    • addedInput schema / properties / ticket
      Added value: +{
      +  "description": "Талон незавершённой задачи из прошлого ответа `[research_pending]`. Передайте его, чтобы дождаться готового результата — бесплатно, без нового списания. / Ticket of an unfinished job from a previous `[research_pending]` response. Pass it to wait for the finished result — free, no new charge.",
      +  "type": "string"
      +}
    • removedInput schema / required
      Removed value: -[
      -  "query"
      -]
  3. First observed

TDQS

A4.5/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Exceptionally rich: pipeline stages, 10–40 s runtime, honest 'incomplete sources' warning plus verify_claim escalation, a documented fallback path when the pipeline is down (3–40 min, `[research_pending]` string with a ticket), and a full pricing model. Annotations already cover readOnly/openWorld, but the description adds far more than they carry. Only gap: no explicit statement of what happens on mid-flight failure of the primary pipeline beyond the fallback.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded well (purpose, when/when-not, mechanics), but it is bloated: the entire content is duplicated in Russian and English, and the 'ticket polling is not recharged' rule is stated three separate times. The complexity justifies length, but the duplication and repetition are avoidable.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

No output schema exists, yet the description fully specifies the return shape (answer text plus source list, with an explicit unverified-warning case), all three execution paths, latency expectations, and cost. Nothing an agent needs to call and interpret this tool is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is already 100%, so the baseline is 3, but the prose adds real invocation semantics: query is required only for a NEW job and may be omitted when polling, and repeating the topic with a new query is a brand-new charge while a ticket poll is free. This shapes how the agent should sequence calls rather than just restating field types.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

States a precise verb chain and resource: it formulates queries, searches, reads sources, and returns a synthesized answer with page links. It is immediately distinguishable from web_search (single fact / link list) and verify_claim (re-checking shaky output), which it names explicitly.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives an explicit 'When' trigger (questions needing several sources reconciled — 'compare', 'analyse', 'what is known about') and an explicit 'When not' with the alternative named (web_search, 20× cheaper, seconds fast). Routing is unambiguous.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources