the knowledge layer for automotive repair

OpenBolts turns fragmented OEM manuals into a single, queryable source of truth. Ask a question, get a verified answer with citations. We’re hoping to work directly with many OEMs to make this knowledge accessible to everyone.

Try the demoView API

The problem

the “knowledge wall” in automotive repair

Modern automotive repair has shifted from mechanical intuition to software-defined precision. But the information infrastructure hasn't kept up. Independent shops are facing a growing crisis.

Data Fragmentation

Repair procedures are trapped in PDFs, clunky OEM portals, and scattered technical service bulletins. There is no single source of truth.

20% of Billable Hours Lost

Technicians waste up to a fifth of their time digging through manuals for torque specs, wiring diagrams, or ADAS calibration steps.

Liability & Safety Risk

Guessing specs in a high-stakes environment leads to safety failures, voided warranties, and massive legal exposure for shops.

Hostile Ecosystems

OEM portals are expensive and structurally designed to favor dealerships over independent shops and small operators.

The US automotive repair market exceeds $100B+ annually. The shops that repair your car are flying blind—navigating critical safety procedures with outdated tools and fragmented data.

Our approach

built for accuracy

OpenBolts is purpose-built knowledge infrastructure for automotive repair, defined by three core principles.

01

Direct Grounding

We own and curate the OEM corpus. Users query the "Truth"—they don't upload their own files. This is a centralized, authoritative knowledge base.

02

Deterministic Accuracy

Every answer includes a page-level citation when from OEM docs; otherwise a web-sourced answer tagged as General. Clear provenance — never invents data.

03

API-First

While we provide a chat UI, our primary value is as a "headless" data layer. Integrate into any shop management software, fleet system, or smart tool.

How it works

tree-guided hybrid search

Each document is parsed into a tree of sections, paragraphs, and procedures. Hybrid search combines structural, semantic, and keyword matching for answers with full provenance.

Phase 1

The Eyes

PDF → structured markdown

  • Preserve headers, tables, and page boundaries.
  • Describe diagrams with vision models (alt-text).
Phase 2

The Brain

Parse into document tree

  • Sections, subsections, paragraphs, tables, figures.
  • Procedure steps as first-class tree nodes.
Phase 3

The Skeleton

Build hierarchical document tree

  • Tree of nodes with page ranges and node types.
  • Each node linked to source doc, page, and metadata.
Phase 4

The Memory

Index for hybrid search

  • Vector embeddings for semantic similarity.
  • Full-text index for keyword search.

Phase 5 — The Agent

runs on every question

  • Parses query intent and optional scope expansion.
  • Hybrid search: structural + semantic + keyword.
  • Traverses tree hierarchy for exact facts.
  • Synthesizes answer with page-level citations.

Answer synthesis with provenance

Produces an answer only when it can cite the exact source. Tiered: OEM-verified first, web-expanded fallback, conversational for non-technical questions.

Why OpenBolts

how we compare

Legacy tools are digital filing cabinets. Generic AI hallucinates. OpenBolts is the retrieval-augmented middle ground: grounded, cited, and purpose-built for automotive repair.

What it is

Legacy Giants

A digital library — folder trees and keyword search through PDFs

OpenBolts

An intelligence layer — semantic understanding of the technician's intent

How you find info

Legacy Giants

Navigate folders, search keywords, scroll through 200-page PDFs

OpenBolts

Ask a natural language question, get the specific fact with its page source

Output

Legacy Giants

A link to a document you still have to read yourself

OpenBolts

The exact answer extracted from the document, with page-level citation

Integration

Legacy Giants

Proprietary portals with no API access

OpenBolts

API-first — embed into any shop management or fleet system

Data source

Generalist AI

General internet training data — no guaranteed OEM provenance

OpenBolts

Directly grounded in ingested OEM PDF manuals only

Accuracy

Generalist AI

Generates plausible-sounding answers that may be completely wrong

OpenBolts

100% provenance — every claim traced to a specific page and document

When it doesn't know

Generalist AI

Hallucinates confidently — invents torque specs, steps, and part numbers

OpenBolts

Returns a web-sourced answer tagged as Web — clear provenance, no invented data

Liability

Generalist AI

No verifiable source chain — unsafe for professional repair

OpenBolts

Full citation trail — auditable, defensible, safe for production use

Integration

one API call. verified answers.

Send a question with optional OEM/model/year filters. Get back a cited answer with document name and page number or a web-based answer when the data isn't in the corpus.

curl -X POST https://openbolts-api.up.railway.app/query \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "question": "How long does the EDR typically record data in a 2023 Toyota Camry?",
    "filters": { "oem": "Toyota", "model": "Camry", "year": 2023 }
  }'

the answer is one query away

Where safety is verified, liability is reduced, and authoritative repair knowledge is always accessible. Try the demo or integrate via API.

Try the demo