What is Keenable and why does it exist now?

Keenable is a San Francisco startup building independent web search infrastructure for AI agents and models. It exists because Google and Microsoft have been restricting third-party access to their search APIs as they consolidate into bundled, selective partnerships. AI builders that previously grounded agents via Bing Search API or Google Custom Search no longer have reliable independent access at scale. Keenable's founders, Andrey Styskin from Yandex and Amazon AGI and Matthias Petri, built the startup to fill that gap with infrastructure designed from scratch for how AI systems retrieve and reason over information.

How does Web Query Language differ from standard RAG or a basic search API call?

Standard retrieval fetches relevant documents and hands them to the model. Web Query Language decomposes a query, retrieves from thousands of sources across the live web, synthesizes results before handing them to the model, and includes provenance metadata for accurate citations. The difference shows up on multi-hop research tasks: comparing data across sources, tracing a timeline, or verifying a claim against multiple conflicting pages. Standard RAG stops at retrieval; WQL reasons over the retrieval step itself.

How does Keenable's MCP integration work in practice?

Keenable ships an official MCP server at api.keenable.ai/mcp. It is keyless by default with a 1,000 request per hour cap. Developers can wire it into Claude Code, Cursor, Windsurf, and any other MCP-compatible client with a single CLI command: keenable configure-mcp --all. The two tools are search_web_pages (live web search with filters for recency, site-specific queries, and date ranges) and fetch_page_content (clean markdown from any indexed URL). Paying customers get an API key that lifts rate limits.

How does Keenable compare to Brave Search API or Exa on price and coverage?

Keenable's Frontier tier prices at $1 per 1,000 requests at 100 RPS+ with optional on-premises deployment, aimed at AI labs running high volumes. Its Agent Builder tier is $4 per 1,000 requests for developers. Coverage is 100 billion documents. Brave Search API indexes around 20 billion pages. Exa focuses on a curated, semantically indexed subset rather than full web breadth. The direct comparisons depend on the use case: Exa is strong for semantic similarity over curated sources, Brave offers independence from big tech, Keenable targets the highest-volume AI lab use case where cost per query and latency at scale are the deciding factors.

Is Keenable replacing consumer search engines?

No. Keenable is infrastructure for AI systems, not an end-user search product. Its customers are AI labs, inference providers, and developers building agents. The relevant comparison is other search API providers, not Google or Bing as end-user products. The company is betting that as AI agents absorb a growing share of web queries (a trend Keenable says is already visible in Cloudflare crawl data), the search index layer will become commoditized developer infrastructure, and the company that owns the most cost-efficient independent index wins.

Keenable Raises $26M to Build Web Search for AI Agents

As of August 25, 2026, AI agents searching the live web have a new infrastructure option. Keenable, a San Francisco startup founded by former Yandex search head Andrey Styskin and AI scientist Matthias Petri, came out of stealth today with $26 million in seed funding led by Accel. The company has built a 100-billion-document search index designed specifically for how AI agents retrieve and reason over information, and its API is already running in production at multiple AI labs and inference providers.

The timing is not accidental. Google and Microsoft have been restricting third-party access to their search APIs throughout 2026, opting for a more bundled, selective partner model. For AI builders that previously used those APIs to give agents live web grounding, the gap is real and growing. Keenable’s seed round, also backed by Conviction Partners and angels from Amazon, ClickHouse, Databricks, Google, Snowflake, and SpaceXAI, is a direct bet that independent web search infrastructure becomes foundational developer tooling in the AI era.

The Gap That Created Keenable

The architecture of most AI agents today has a structural weakness: models train on data with a knowledge cutoff, so anything that happened after that cutoff requires a live retrieval step. That retrieval step typically relies on one of three paths: a proprietary search API from Google or Microsoft, a specialized semantic index like Exa, or a company managing its own crawl infrastructure, which is prohibitively expensive at web scale.

Path one is closing down. Path three is out of reach for most teams. Path two has strong semantics but limited breadth. Keenable is competing to own path four: an independent, cost-efficient, full-web index built from scratch for AI-scale query patterns.

Styskin described the core technical insight in the TechCrunch interview: “If you do not fine-tune your index structures for a specific task, the cost of serving and scanning the whole internet is enormous because of the volume. That’s why you need to innovate on how you can narrow the search space based on your query very fast. This is what we are bringing to the table.”

That is not a trivial claim. Styskin spent 20 years building search at Yandex and then Amazon AGI, where he and Petri built the web grounding service behind Alexa. They have built web-scale search indexes twice before. Keenable is their third.

What Keenable Is Actually Shipping

Keenable has two products in production today and one in development.

The first is a Search API over its 100-billion-document index. The index supports standard query parameters plus AI-specific filters: recency windows, site-specific scoping, and date-range constraints. Latency is under 250 milliseconds at the 95th percentile from US East. The service is already running in production at several AI labs and inference providers, though Keenable has not disclosed customer names.

The second is an official MCP server at api.keenable.ai/mcp. It is keyless by default with 1,000 requests per hour, making it the fastest path to adding live web access to an agent without an account. The server exposes two tools: search_web_pages and fetch_page_content. The CLI command keenable configure-mcp --all detects and configures Claude Code, Cursor, Windsurf, and other compatible clients automatically. This matters for enterprise developers because it makes Keenable a drop-in addition to existing agent toolchains, not a platform migration.

The third product, Web Query Language (WQL), is in development. WQL targets multi-source synthesis: questions where the full answer cannot be found on any single page. It decomposes a query, retrieves across thousands of sources, synthesizes the result before the model sees it, and returns provenance metadata for citations. Enterprise AI teams building research agents, competitive intelligence tools, or due diligence workflows will find WQL directly relevant once it ships.

Pricing and Access Model

Keenable operates on a tiered pricing structure that scales with volume:

Tier	Deployment	Price	Target
Keyless	Cloud only	1,000 req/hr free	Prototyping
Agent Builder	Cloud only	$4 / 1K requests	Developers, small agents
Frontier	Cloud + on-premises	$1 / 1K requests at 100 RPS+	AI labs, inference providers

The Frontier tier on-premises option is significant for regulated enterprise buyers who cannot send queries through a third-party cloud. It is also a competitive moat: building an on-premises option means maintaining an index delivery format, a deployment package, and a support model that most search API startups skip. The angel cap table, which includes investors from ClickHouse, Databricks, and Snowflake, suggests Keenable is already thinking seriously about enterprise data infrastructure partners.

The Competitive Landscape

Keenable enters a market that already has several players, but none with exactly the same positioning:

Brave Search API: Independent of Google and Microsoft, approximately 20 billion page index, strong on privacy. Positioned more for general search than AI-scale infrastructure.
Exa: Semantically indexed, optimized for “find pages like this” queries. Strong for curated semantic retrieval; narrower breadth than a full web crawl.
Tavily: Search API built for RAG pipelines. Focused on clean retrieval output rather than index scale.
Google and Microsoft: Restricting third-party API access in 2026. Still usable through partner programs, but reliability for independent AI builders has declined.

Keenable’s differentiation is primarily scale and purpose-built architecture. It is not trying to build the best semantic index. It is trying to build the most cost-efficient full-web index for AI systems that need raw coverage: training data retrieval, agent web grounding, runtime fact-checking across the live web.

Accel partner Zhenya Loginov, who led the investment, made the competitive logic explicit: “AI players have very few options when it comes to web-scale search infrastructure, especially with Google and Microsoft taking steps to shut down their existing search APIs. Keenable’s conviction is also shared at the Googleplex: whether it’s for humans or for agents, the era of the ten blue links may be coming to a close.”

What Enterprise AI Teams Should Do Now

For enterprise teams building agents that need live web data, Keenable is worth evaluating immediately for three reasons.

First, the keyless tier removes the friction from a first test. A developer can wire the MCP server into Claude Code or Cursor in under five minutes, run queries against the live index, and see actual results before signing up for anything.

Second, the Frontier tier’s on-premises option addresses the data residency and compliance questions that block most external API adoption in regulated industries. If the index can run on enterprise-controlled infrastructure, the standard objections about query logging and data sovereignty disappear.

Third, the WQL product, once it ships, directly enables the multi-source synthesis workflows that are hardest to build today. The enterprise use cases that need it most, including AI-powered legal and financial research, due diligence, and competitive intelligence, are also the ones with the clearest willingness to pay.

The deeper strategic point is architectural. As more enterprise AI agent infrastructure becomes available as independent, MCP-native components, the cost of assembling a capable agent stack continues to fall. An agent that can search the live web via Keenable, retrieve structured knowledge via Pinecone Nexus, and act through a capable foundation model covers a large share of knowledge-work use cases without locking into any single vendor’s bundled platform.

The bundled AI platform vendors want enterprise buyers to believe that integration is their moat. Keenable’s arrival, and the broader pattern of specialized infrastructure startups filling each layer of the agent stack, is evidence that the unbundled path remains viable, and in many cases cheaper, for teams willing to assemble it themselves.

Keenable’s index is live at keenable.ai. The MCP server is available now. WQL is in development without a public timeline. The team of 15 plans to double by end of year, using the seed capital to expand the index, crawling infrastructure, and the retrieval technology underlying WQL. Book a call with Enera if you are evaluating independent web search infrastructure for your agent stack and want to work through the architecture.