AnySearch

AnySearch is search infrastructure built for AI agents rather than a human-facing search box. The current product exposes a REST API, MCP integration and installable agent Skill and routes queries across web and specialized sources before returning filtered, de-duplicated and structured results.¹²

Its core thesis is that agents need a different retrieval layer from human search. A person can skim duplicate links and decide what to trust; an agent pays for every token and may reason confidently over stale, repetitive or poorly sourced context. AnySearch therefore tries to handle routing, source selection, redundancy reduction and result structure before context reaches the model.¹

AiToolMap evaluates AnySearch as retrieval infrastructure, not as an LLM. The relevant questions are source quality, de-duplication, provenance, integration, latency, privacy and operational maturity.

The current API exposes `/v1/search` and `/v1/deep-search`.²⁷ Search requests can specify the query, result count, domain/sub-domain tags, geographic zone, language and source-specific parameters, with JSON or Markdown output.²

The point is that financial data, technical documentation, academic material and general web pages have different authority, freshness and structure requirements. AnySearch tries to route the query to suitable sources rather than treat every request as the same generic web search.

Product Hunt maker discussions describe source-aware freshness and caching strategies rather than one global TTL.³ That remains vendor evidence, but it is consistent with the product design.

AnySearch markets structured output heavily. Current documentation guarantees a consistent result structure with source metadata and JSON/Markdown output.²

Product Hunt discussions clarify that arbitrary caller-defined strict schemas are not the central current behavior.³ The strength is predictable retrieval packaging rather than universal structured extraction.

For agents, that is still valuable: less HTML cleanup, fewer duplicate passages and explicit provenance mean more of the context window can be spent on useful evidence.

The Terms define Deep Search as an LLM-based multi-turn search service.⁷

This can reduce orchestration work for agents that need to break a question into multiple searches. Developers who already have their own planning/research layer may prefer the lower-level Search endpoint; teams wanting retrieval plus research decomposition can test Deep Search as a higher-level primitive.

No large independent benchmark separating the quality of Deep Search's reasoning from the underlying retrieval layer was found in the fixed panel.

AnySearch supports API, MCP and agent Skill distribution.¹² This fits the way current coding/research agents consume tools and reduces the need to build a bespoke wrapper before evaluating the service.

Secondary ecosystem reporting also shows early developer interest around these distribution paths.⁸ Those adoption signals are not satisfaction ratings, but they support the view that AnySearch is aimed at agent infrastructure rather than consumer search.

The API documentation supports anonymous requests as well as Bearer-key authentication.²

The Privacy Policy says full API request/result Business Data is not persistently stored after the response and temporary processing data is destroyed after processing.⁶ That is a meaningful design choice because search queries can reveal a user's or company's research interests.

Anonymous mode also lowers evaluation friction: developers can test the service without creating a full account flow.

The Privacy Policy gives a more precise operational picture than the homepage slogan.⁶

For authenticated API usage, AnySearch records metadata such as request time, API-key identifier, search type, routing domain, response status, latency and result count. It also retains the first 100 characters of a query as a Search Query Summary for troubleshooting/service improvement.⁶

The company says it does not perform semantic profiling on that query content and does not persist the full request/result payload after the response. API logs are retained for 12 months.⁶

This remains privacy-forward relative to storing a complete search history, but “No Logging” should not be read literally as “no operational records of any kind.” Enterprise buyers should rely on the policy and contract language rather than the marketing shorthand.

Product Hunt maker responses describe Private Capability Isolation for Enterprise users, allowing search to be constrained to company documents or a whitelisted source set.³

That could be important for support or regulated agents where open-web evidence is undesirable. AiToolMap treats it as vendor-described functionality rather than independently tested policy enforcement; teams should verify allowlists and failure behavior contractually.

Free costs $0 and provides 1,000 requests per day, structured output and 20 QPS per API key.⁵

Search Pro is currently Coming Soon. The public page describes higher-quality search, larger quotas, higher rate limits and deeper vertical capabilities but does not publish a price.⁵

Enterprise is custom and adds tailored limits, security/privacy/compliance controls, advanced vertical search, dedicated account management and priority support.⁵

The practical consequence is that AnySearch is very easy to prototype today, but there is not yet a mature public self-serve paid middle tier to compare by price.

AnySearch's own benchmark uses 300 questions across Frames, FreshQA and WebWalkerQA while holding the model fixed at `z-ai/glm-5.1` and changing the search provider.¹

The company reports 76.4% overall accuracy, 80.0% on FreshQA and 65.2% on WebWalkerQA, with 47.8 seconds average end-to-end latency overall. Compared systems are reported as slower in the tested configurations.¹

The design is more useful than a vague “better search” claim because it attempts to isolate retrieval. But the benchmark was designed, executed and published by AnySearch, so AiToolMap does not normalize it as an independent score.

AnySearch searches public and third-party sources. Its Terms state that third-party result content is not guaranteed for legality, accuracy, completeness, stability or intellectual-property status.⁷

The Privacy Policy describes customers as Data Controllers for personal data retrieved through search while AnySearch acts as processor for the service.⁶

That matters for regulated screening, legal research, due diligence or customer-facing agents. Retrieval infrastructure does not remove the caller's duty to establish lawful use, verify critical facts and apply source/human-review controls.

Product Hunt currently shows 4.0/5 from one review.³ The qualitative content is more useful than the average: easy integration, stronger information density and a request for debug visibility into ranking decisions.

Hunted.Space captured 569 launch upvotes and 117 comments and a #1 Product of the Day result.⁹ Product Hunt also shows substantial launch attention.³

Those are adoption signals, not satisfaction metrics, and they are excluded from the future rating equation.

The Privacy Policy says passwords use bcrypt, API keys are hashed and data in transit uses HTTPS/TLS.⁶

The homepage also advertises zero-knowledge credentials and private capability isolation.¹ Enterprise promises enhanced security/privacy/compliance support.⁵

AiToolMap did not independently verify a current SOC 2 or ISO certification report in the fixed panel, so no such certification is claimed here.

The strongest verified trust feature is narrower and concrete: full API query/result payloads are not persistently retained after execution according to the policy.⁶

AnySearch is most relevant to agent developers who already see duplicate context, weak provenance or excessive cleanup from generic web-search APIs.

It is particularly attractive for MCP/Skill-based research and coding agents because integration friction is low and the free quota is large enough to evaluate on real workflows.

Teams building customer-facing or regulated agents should investigate Enterprise source isolation and contractual logging/data controls before production use.

Human users who simply want a search website are not the target audience. AnySearch is infrastructure intended to sit inside another AI system.