Bixel

TopK

Topk.io
Company Overview

As of September 3, 2026: TopK is a usage-based search engine for AI applications that positions against Bedrock Knowledge Base and Gemini File Search on a claim of 10x lower cost backed by object-storage architecture.

The service charges across nine separate usage dimensions, from $0.005 per TiB scanned for query compute to $20 per GiB for query memory, so cost predictability depends heavily on workload shape. A $5.5M Seed round funds a two-person engineering hiring push, and the company holds SOC 2 Type I but not yet Type II.

Backed by 48 dated facts
Drawn from 8 dimensions of the record, each dated and sourced.
open roles
2
Sep 2026
technologies
9
observed
dated changes
101
Aug 2026
The record, by dimensionevery dimension is its own page →
PricingBacked by 10 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.
Usage-based
Nine metered dimensions with no flat tiers means there is no monthly ceiling: a memory-heavy query workload hits $20 per GiB, while storage runs $0.10 per GiB, so buyers should model their exact access patterns before committing.
Full pricing read →
Features
Hybrid search platform
The platform states hybrid search and multi-tenancy across 9 capabilities with up to 10 namespaces; buyers building multi-tenant AI apps get namespace isolation out of the box, but the capability list is self-reported and no third-party benchmark is cited.
Backed by 10 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Security
Baseline security
SOC 2 Type I is in hand with encryption at rest, encryption in transit, and private networking confirmed; Type II is not yet listed, which matters for enterprise procurement teams that require the audit-period evidence Type II provides.
Backed by 5 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Reliability
Operational
A public status page exists and currently reads operational, but no published uptime SLA or historical incident log was observed, so buyers cannot verify the reliability claim against actual track record.
Backed by 2 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Hiring
2 open roles
Two open roles, both in engineering, with no remote policy stated; at this headcount level, a buyer should factor in that core product and support capacity is thin.
Backed by 3 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Stack
Modern web stack
The live site runs on AWS and Cloudflare with Datadog for observability and Next.js on the front end, plus five additional detected technologies; the AWS and Cloudflare pairing is consistent with the object-storage, global-edge architecture the homepage describes.
Backed by 10 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Positioning
Search Engine
The homepage headline is "Store anything. Find everything." with an explicit cost claim of 10x lower than Bedrock Knowledge Base and Gemini File Search; that claim is stated by the company and is not independently verified here.
Backed by 6 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Funding
$5.5M raised
A single $5.5M Seed round is the full capital base on record; buyers and investors should note this is early-stage runway, and the company has not yet demonstrated a Series A raise.
Backed by 2 dated facts
Every value derives from a dated, sourced capture — open any fact for its source.Full read →
Change History101 dated events on the record
Feature removed: Postgres · high significance
Feature removed: Late Interaction · high significance
Feature added: Namespaces
Feature removed: Dense Vectors · high significance
Feature removed: Document Processing · high significance
Feature removed: File Search · high significance
Feature removed: Ingest Throughput · high significance
Get this record via API / MCPevery field dated and sourced
RESTopen tier
GET https://api.bixel.com/v1/companies/topk.io

{
  "data": {
    "company": {
      "name": "TopK",
      "domain": "topk.io",
      "categories": [
        "vector-databases"
      ]
    },
    "dimensions": {
      "security": {
        "facts": [
          {
            "key": "security.cert.soc2-type-i",
            "value": true,
            "provenance": "company_stated",
            "as_of": "2026-09-03",
            "source_url": "docs.topk.io"
          },
          {
            "key": "hiring.open_roles",
            "value": 2,
            "provenance": "company_stated",
            "as_of": "2026-09-03",
            "source_url": "www.topk.io/careers"
          },
          {
            "key": "stack.tech.aws",
            "value": true,
            "provenance": "observed",
            "as_of": "2026-09-03",
            "source_url": "trust.topk.io"
          }
        ]
      },
      "…": "…"
    }
  }
}
MCPfor agents
# any MCP client (Claude, agents)
const record = await bixel.get_company({ domain: "topk.io" })

# returns the whole record above,
# each value with its source_url + as_of,
# ready to reason over
Build on the company record. One key, REST + MCP, every signal dated and sourced back to the page it came from.
How Bixel reads this

Every read above is derived from public signals, each sourced and dated, and kept honest about provenance: what the company states on its own pages and job posts (pricing, security, careers, positioning, its backend stack) versus what Bixel independently detects (technologies, infrastructure). Where a stated claim is also detected we mark it verified; where we only have the claim, we say so. Bixel infers posture: how it monetizes, how mature it is, where it's heading. It does notclaim private financials it can't observe. Where signals are thin, the record says so.

Public record, read from companies' own pages and boards. Every fact dated and sourced; provenance (observed vs company stated) shown inline.