Skip to content

GTools · Building useful AI

AI thatshows its work.

GTools builds useful AI for real businesses: support that cites its sources, a gateway that accounts for every token and memory that returns its evidence.

AI customer support, Grounded answers, One script tag, LLM gateway, Agent observability, Cost attribution, Hybrid retrieval, Knowledge graphs, Graph RAG

// 01Manifesto

Most AI guesses. Ours is built to answer from what you actually know, to show the source behind every reply, to say “I don't know” when it doesn't, and to account for every token it spends. That's what we mean by useful.

// 02Products

Three products.
Each one useful on day one.

01/03AI customer support

Pingie

Answers your customers from what you know. Only that.

Train Pingie on your files, pages and team answers, then add it to your site with one script tag. Every reply shows its source. When Pingie doesn't know, it says so and hands the visitor to a person.

  • 01One-line script install
  • 02Cited answers from your own sources
  • 03Honest fallback to a live person
  • 04Missed questions become leads and knowledge gaps
pingie · widget
Online

Pingie

Answers only from your sources

Ask a question…
<!-- one line to install -->
<script src="…/widget.js" data-bot-id="bot_…" defer></script>

02/03LLM gateway · Agent observability

Vizhi

See every agent, every model call, every token.

Connect your model providers once, issue model tokens and agent tokens, and route traffic through one gateway. Vizhi shows usage, latency, errors, tokens and estimated cost for every agent and every model.

  • 01One gateway for every LLM provider
  • 02Separate model and agent tokens
  • 03Live usage, latency and error metrics
  • 04Cost attribution per agent
vizhi · overview
Live

Requests / min

427

p95 latency

840 ms

Tokens today

1,842,310

Est. cost

$18.42

Agents, illustrative data
AgentCallsLatency
support-bot1,284
research-agent342
billing-agent96
triage-agent518

03/03Memory for AI applications

Memora

Durable, retrievable memory that brings receipts.

One API to store files, URLs and chats and ask questions about them. Memora retrieves two ways: vector RAG over structural chunks, and graph RAG over a knowledge graph, on their own or combined. Every answer comes with the exact chunks, facts and scores behind it.

  • 01Vector RAG: dense + lexical, RRF, MMR
  • 02Graph RAG: PageRank, paths, communities
  • 03Structural chunks with an HNSW index
  • 04Grounded answers with evidence
memora · POST /api/v1/query
Illustrative
What is the refund window for annual plans?

Vector RAG

  1. Structural chunkssection-aware · 300–500 tok
  2. Dense + lexicalHNSW · Postgres full-text
  3. RRF fusionreciprocal rank · k = 60
  4. MMRdiversity · λ = 0.7

Graph RAG

  1. Knowledge graphentities · relations · facts
  2. PageRankpersonalized to the query
  3. Pathsmulti-hop between entities
  4. Communitiesclustered context

Grounded answer

Annual plans can be refunded within 30 days of purchase.

chunk · Refunds › Annual plans · p. 2similarity 0.81fact · Annual plan → refund window → 30 days

// 03Ecosystem

Three products.
One standard.

Each GTools product does one job well and holds the same line: answers you can trace back to a source, systems you can see into, and data that stays with its owner.

  • 01PingieTalks to your customers
  • 02VizhiWatches your agents and models
  • 03MemoraRemembers what your apps know
GTools
Pingie
Vizhi
Memora

// 04Principles

What useful
means to us.

Four lines every GTools product holds, taken from how they already work.

01Seen in · Pingie · Memora

Grounded

Answers come from your own material, with the source attached. If it isn't in your data, it isn't in the answer.

02Seen in · Pingie

Honest

When the answer isn't there, our software says so and hands over to a person instead of guessing.

03Seen in · Vizhi · Memora

Observable

Every model call is counted: tokens, latency, errors and cost, per agent and per request.

04Seen in · Memora

Private by design

Every request is scoped to its owner. Another tenant's data isn't hidden from you; it's unreachable, down to a 404 rather than a 403.

// 05Contact

Let's buildwhat's useful.

Try a product, ask about early access, or tell us what you're building.