Keenable Review

Own crawler, own index, own ranking. That's the real claim here, not the markdown formatting or the MCP server — most search APIs for agents are thin wrappers on someone else's results, and Keenable says it isn't one.

Pricing is concrete: free keyless up to 1,000 req/hour, 100K free monthly after signup, then $4/1000 requests, dropping to $1/1000 at 100 RPS+ for labs wanting dedicated or on-prem capacity. That's a real ladder from hobbyist to infrastructure buyer, not a single contact-sales wall.

Tradeoff: an independent index is a hard, slow thing to build, and I can't verify ranking quality or coverage from here. The Time Machine point-in-time search is a genuinely distinct feature if it works as described. Pilot it against your current provider on real queries before committing spend.

Most 'search API for agents' products are a thin client over Bing or a scraping layer with caching. Keenable claims its own crawler, index, and ranking system end to end. That's a meaningfully different build — it means latency, freshness, and coverage are theirs to fix or break, not a pass-through SLA from an upstream vendor you don't control.

The AgentID programmatic key issuance and the MCP server with one-click Cursor/VS Code installs tell me they designed for agent-initiated onboarding, not human dashboard flows — that's the right shape for 2026 agent infrastructure. Time Machine's query_time parameter for point-in-time corpus rewinds is a genuinely unusual primitive; most search APIs give you now, not provable history.

The tradeoff: at $1-4 per 1,000 requests with Dedicated Capacity gated behind contact-sales, I can't model true cost-at-scale until I'm already talking to their team. Three years in, if the index keeps pace with the open web, this is a defensible primitive to build on. If it doesn't, you're stuck migrating a crawler dependency, not swapping an API key.

100,000 free requests a month. Then $4 per 1,000 on Agent Builder, $1 per 1,000 at 100 RPS+ on Frontier. Usage-based, no seat math, no SSO tax to budget around.

Run the numbers: an agent doing 500K calls a month burns through the free allowance fast, then it's 400K billable at $4/1000 — $1,600/month, $19,200/year. Move that same volume to Frontier's $1/1000 and it's $400/month if you clear the RPS bar. Tier choice is the real lever here, not negotiation.

No stated contract term, no auto-renewal clause I could find — usage-based pricing usually means month-to-month, cancel by stopping calls. That's a procurement plus: no multi-year lock-in to unwind. Downside: Pay as you go and Dedicated Capacity both say 'contact sales' for final pricing, so the published per-1000 figures may not be the number on your invoice. Get that in writing before you model year 3.

Page Fetch returning clean markdown instead of raw HTML is the detail that matters day-to-day — that's one less BeautifulSoup/Readability pass in your agent's tool chain, and one less place for silent extraction bugs to hide. Ship that as a first-class API response and you've removed a chunk of glue code most teams write themselves.

The MCP server plus Homebrew CLI means you're not writing a bespoke HTTP client to get this into Cursor or Claude Code — one-click install, done. That's the right ergonomic bet for 2026 agent tooling, where the client ecosystem fragmentation (Cursor, Claude Desktop, Codex, VS Code) is itself the integration tax.

AgentID for programmatic key issuance is interesting for anything spinning up ephemeral agents at runtime, but I'd want to see rate-limit and auth-failure behavior under real concurrency before trusting it in a production fleet. The tradeoff: no per-call usage visibility until you create an account removes the ability to debug cost or volume anomalies during that first keyless exploration window.

Most APIs make you create an account, grab a key, paste it somewhere, then find out if the thing even works. Keenable skips that — point an MCP client at the endpoint and you're searching. That's a genuinely different first ten minutes than category norm, and it matters more than people admit, because the gap between 'signed up' and 'saw a real result' is where most devs quietly give up.

The CLI via Homebrew and one-click installs for Cursor and VS Code suggest someone thought about the terminal person and the IDE person separately, which is its own small kindness. Clean markdown instead of raw HTML on page-fetch is the kind of detail that saves you a parsing step you'd otherwise build badly yourself.

Where I get cautious is month three. A 100,000-request free allowance is generous for testing, but I have no feel for how the dashboard behaves under real volume, what error states look like at 100 RPS, or how support responds when billing gets weird. Those are the things that decide if you renew, not the demo.

"Independent crawler, index, and ranking" is the whole differentiation story. Fine claim. Also the one I've heard from search wrappers that turned out to be thin reranking layers over someone else's results once you pushed on latency or coverage at scale. Keenable says ranking improves as more agents use it — plausible, unverifiable, the kind of self-reinforcing claim that's either true or a nice story.

Track record is otherwise concrete: $4 per 1,000 requests on the builder tier, $1 per 1,000 at 100+ RPS for Frontier, 100K free monthly requests. Real numbers, easy to model.

AgentID programmatic key issuance is a nice touch for fleets of agents spinning up fast. Exit risk is the standard API-dependency one — swap the endpoint, rebuild whatever ranking assumptions you baked in. Two sales-gated tiers above the published rate card is a normal SaaS move, not a red flag by itself.