Keenable Review

1,000 keyless requests an hour, no key, no config file. That's a real try-before-you-commit move, not a marketing gimmick. Free accounts jump to 100,000 requests a month with per-call tracking, then $4 per 1,000 requests pay-as-you-go, or $1 per 1,000 at 100 RPS+ for dedicated capacity.

Own crawler, own index, own ranking — that's the thing that matters here. Anyone can wrap another provider's search API. Building the index is the hard, slow, expensive part, and the one that actually differentiates from a thin reseller wrapper.

Time Machine, a query_time parameter that searches prior page versions, is a genuinely unusual capability I haven't seen elsewhere in this exact form. Tradeoff: cloud-only at the entry tier, on-prem only shows up once you're negotiating dedicated capacity.

Most 'search for agents' products are a thin billing layer over Bing or a scraper farm. Keenable's pitch — its own crawler, index, and ranking — is a different dependency graph entirely. If that holds up under load, you're not exposed to a reseller's rate-limit changes or a shutdown two years out; if it doesn't scale, you own an unproven index instead of a mature one, and that's a real bet either way.

The integration surface is unusually considered for a v1: MCP server, Homebrew CLI, one-click Cursor/VS Code installs, plus documented LangChain, LlamaIndex, and Haystack paths. AgentID letting an agent provision its own key programmatically is the detail that tells me they've thought about machine-to-machine auth, not just human developer onboarding.

The tradeoff shows up in the pricing architecture: $4/1000 requests cloud-only on the entry tier, $1/1000 at 100 RPS+ only on the dedicated tier, and on-prem gated behind a sales conversation. That's a normal shape for infra pricing, but it means your cost curve and your deployment options move together — plan your 3-year spend assuming you'll eventually need that Frontier tier.

$4 per 1,000 requests, pay-as-you-go, cloud-only. 100K free requests a month first. A team burning 2M requests/month lands near $7,600/year at the entry rate — before touching the $1/1,000 Frontier tier, which needs 100 RPS+ and a sales call to price.

That's the split worth flagging. Agent Builder tier is a rate card. Dedicated Capacity is quote-only. Procurement can self-serve the first, not the second.

Billing is usage-based, no seat math, no SSO tax to budget around. TCO scales with call volume, not headcount — cleaner than most per-seat tools, harder to forecast if traffic is spiky. No term length or auto-renewal language surfaced for the paid tiers, which is the real open question before signing anything beyond pay-as-you-go.

First thing I check on any search API: does it return raw HTML I have to parse, or something an LLM context window can actually use. Page fetch returns clean markdown, which cuts out a whole preprocessing step most teams end up writing themselves against other search APIs. That's a real day-3 win — one less regex-laden scraper module to maintain.

The CLI installs via Homebrew and handles MCP config directly, so bootstrapping a Cursor or VS Code connection doesn't mean hand-editing a JSON config file. AgentID letting an agent provision its own key programmatically is a nice touch for anything spinning up ephemeral agents rather than a human clicking through a dashboard once.

What I can't tell from docs alone is ranking consistency under load — an independent index is a bigger variable than a wrapper over an established backend, and that only shows up after weeks of production queries, not a demo. Framework docs cover LangChain, LlamaIndex and Haystack, which is the right integration surface for 2026 agent stacks.

There's no UI to speak of here, which is honestly the right call for something billed as infrastructure. The Homebrew CLI and MCP one-click installs for Cursor and VS Code mean your first ten minutes is running a command, not clicking through a tour. That's a good sign for a dev tool — it respects that the target user already lives in a terminal.

The API Dashboard is where daily polish actually gets tested, since it's the one screen a human looks at. Usage-per-call tracking kicks in once you make a free account, which is the right tradeoff — you get 100,000 requests a month to figure out if the thing even works before worrying about billing precision.

Where I'd want more before betting a production agent on it: reliability under load isn't something you can see from a dashboard screenshot, and there's no mobile story at all, though for this category that's a non-issue, not a gap. The learning curve is shallow at first, steeper once you're tuning ranking behavior for real traffic.

Independent index. That's the whole marketing hook here — not reselling Bing or Google results, building the ranking yourself. Bold claim. Also the exact claim every search-infra startup makes before quietly falling back to a wrapped API when crawl costs bite. Nothing here proves it's real. Nothing here proves it isn't. Worth watching.

The two-tier split — $4/1000 requests cloud-only, $1/1000 requests at 100 RPS+ dedicated — is transparent pricing, credit where due. But 'contact sales' sits behind both real tiers once you're past the free 100K. That's not hidden pricing, it's a floor you don't know until you're already a customer.

AgentID and the MCP server show real integration thinking. Migration off a self-built index, though, means re-pointing to whatever alternative you'd have used anyway — Bing API, Exa, whoever. Nothing here locks you in. Nothing here locks you to stay, either.