Launch watch · 25 August 2026

Keenable’s launch has integration traction, but little independent launch-day reaction

Across 2,300+ unique search results scanned, most launch-day coverage repeated the funding and index story. The more useful outside signal is in code and benchmarks: several integrations landed, two clients exposed the same result-field mismatch, and independent tests show a fast, inexpensive general-web product with real specialist-search gaps.

$26m
seed round reported
2
same schema mismatch seen
3+
upstream integrations merged
33.7%
publication R@10 in one test
Where the useful outside signal is
selected evidence items · not total web mentions
Benchmarks
4
Integration work
4
Concrete bugs
2
Original press
1

A curated evidence set from the search, weighted toward sources that added information rather than repeating the announcement.

01

The clearest bug is an API-contract mismatch, not a search outage. Vellum rendered title, URL and date but no result text because it read a field Keenable usually leaves empty; Ketch and Omnigent fixes explicitly switched from description to snippet. docs.keenable.ai

02

Adoption is real, although much of the integration push appears vendor-assisted. Nanobot and CrewAI merged providers, Osaurus added Keenable to its remote MCP catalog, and Hermes Web Search Plus added keyed and opt-in keyless modes. github.com

03

Outside tests support the speed-and-cost pitch, but not universal retrieval superiority. A reviewed extraction test put uncapped /v1/fetch second of eight at 0.862 recall, while an academic-publication test put Keenable at 33.7% R@10 versus Exa at 73.3%. krabarena.com

04

Benchmark claims need version and disclosure checks. One public 24-question claim initially reported 24/24, then its verification timeline recorded a 23/24 rerun. Other Keenable-favorable KrabArena tests explicitly disclose affiliation or remain unreviewed. krabarena.com

Three things worth watching next

API reliability

Schema consistency

Whether SDKs converge on one documented text field and whether compatibility tests prevent blank snippets across clients.

Organic demand

Who authors integrations

Several visible pull requests come from Keenable-associated contributors. Maintainer-led additions and user issues would be stronger adoption evidence.

Product boundary

Specialist indexes

General web freshness and latency look competitive; publications and other specialist corpora are where dedicated providers may retain a material edge.

Evidence shortlist

SignalWhat it saysInterpretationSource
Launch pressTechCrunch reported the launch; many later pages repeated the same $26m/indexing narrative.Visibility, but little reaction yettechcrunch.com
Blank textVellum results show metadata without text because of an empty field.Concrete integration bugdocs.keenable.ai
Field fixKetch changes result text from description to snippet.Same contract issuegithub.com
Field fixOmnigent makes the same snippet-not-description correction.Cross-client patterngithub.com
IntegrationHermes adds authenticated endpoints and opt-in public access; public tier has no SLA.Useful adoption and risk framinggithub.com
ValidationCrewAI added a check rejecting negative n_results.Small client-side edge casegithub.com
Full-page fetchUncapped fetch recall measured 0.862; an older 50k cap truncated long pages.Strong result, cap matteredkrabarena.com
PublicationsKeenable scored 33.7% R@10 versus Exa’s 73.3% on a 300-query subset.Material specialist-search gapkrabarena.com
LeaderboardArtificial Analysis lists pro and realtime variants across quality, latency and cost.Independent comparison surfaceartificialanalysis.ai

Method: searched 18 query angles across 2,300+ unique results on 25 August 2026, then reviewed selected launch coverage, GitHub changes, integration documentation and benchmark pages. Counts in the chart describe the curated evidence shortlist, not all mentions. Company and investor posts were excluded from the conclusions; affiliated or unreviewed benchmarks are labelled as such.

Keenable · made with SELECT* · 1,839 pages in 1m 55s · Open the chatShare: XLinkedInReddit