0
Training retention
Customer corpora are not pooled for model training
Private preview · zero retention · military-grade privacy
Super private. Extremely accurate. Engineering AI that respects your standards.
Zero retention. Multi-provider routing — OpenAI, Claude, Gemini, Groq, Kimi. Five-plus turbo engines. Verification loops. Vevesh indexes the folders you already trust — then answers with citations your lead can sign.
Your folders stay local
Citation-grounded answers
Team knowledge bases
vevesh://workspace/turbine-standards
You asked
What are the fatigue load combination requirements for an IEC Class IIA site with 50-year design life?
Vevesh answer
GroundedFor Class IIA, your indexed IEC 61400-1 section 7.4.2 requires DLC 1.2 and 6.4 combinations with the site-specific turbulence class applied. Your foundation report (Rev C) already flags the governing combination as My at tower base under DLC 6.1...
iec61400-1.pdf · p.84
foundation-revC.docx · §4.2
0
Retention training
Zero retention on customer corpora
5+
Turbo engines
Express → Premium extraction
100%
Citation grounded
Evidence-ready answers
3
Surfaces
Phone · desktop · browser
Military-grade privacy Local + cloud hybrid Verification loops Conflict resolution Global sector KBs Consultant mode Team RBAC Offline & online RAG Live sync review Tool calling at scale PhD research engine File mesh visualization
// OpenAI GPT · Anthropic Claude // Google Gemini · Groq inference // Moonshot AI Kimi · DeepSeek // xAI Grok · Alibaba Qwen // Cloudflare Workers AI gateway // LangGraph tool calling // Engineering RAG · folder-native // Cursor AI workflow compatible // PDF & Office extraction // IEC / ISO standards citations // Hybrid vector + BM25 search // Zero retention privacy // Team knowledge bases · RBAC // ChatGPT alternative for engineers // Local RAG · self-hosted path // OpenAI GPT · Anthropic Claude // Google Gemini · Groq inference // Moonshot AI Kimi · DeepSeek // xAI Grok · Alibaba Qwen // Cloudflare Workers AI gateway // LangGraph tool calling // Engineering RAG · folder-native // Cursor AI workflow compatible // PDF & Office extraction // IEC / ISO standards citations // Hybrid vector + BM25 search // Zero retention privacy // Team knowledge bases · RBAC // ChatGPT alternative for engineers // Local RAG · self-hosted path
Super private · zero retention · military-grade boundaries
Zero retention privacy 5+ turbo engines Military-grade boundaries Local · cloud · mobile · desktop Standards-grade citations Verification loops
Privacy & trust
Zero retention.
Your knowledge is not our product.
Consumer AI treats your uploads as fuel. Vevesh treats them as classified engineering material — isolated planes, explicit boundaries, and architecture that assumes your worst-case leak is unacceptable.
3
Isolated CF accounts
Marketing · ops workers · user KB data — separated
∞
Local-first option
Index on your machine; cloud only when you choose
RBAC
Team privileges
Customizable access per project and knowledge base
The problem
Your knowledge is already organized.
Your AI tools pretend it is not.
Decades of specs, reports, and calculations sit in directories your team trusts. Then someone asks you to drag those files into a chat window — again — and trust a summary with no page number.
- 1 Upload limits and session resets erase context between reviews.
- 2 Wiki migrations fail because engineers never live in wikis — they live in folders.
- 3 Enterprise search connects SaaS apps but not your mixed PDF/CAD/code project tree.
The Vevesh answer
Index where the work already lives.
Persistent folder indexes. Four extraction engines. Citations your lead engineer can verify before sign-off. Team knowledge bases when the corpus belongs to everyone — not just the laptop that indexed it first.
See how we comparePlatform capabilities
Everything serious teams asked for — built in
Privacy, turbo engines, verification loops, team RBAC — not a slide deck. Explore the full stack on the features page.
Privacy & trust
Super private. Zero retention. Military-grade boundaries.
Vevesh is architected so customer corpora stay in isolated data planes — not pooled into shared model fine-tuning. Query traffic is processed for the answer, not harvested for product analytics.
All privacy features →Core
Zero retention by design
Your prompts and source files are not a training dataset.
Enterprise
Military-grade privacy posture
Boundary-first engineering for teams who cannot leak load cases.
Hybrid
Local AND cloud — you choose
Desktop-native indexing or edge pipeline — same folder authority.
Multi-surface
Phone · computer · browser
Field engineers on mobile. Leads on desktop. Reviewers in the browser.
5+ engines · Verification loops · Evidence reporting · File mesh · Consultant mode · Offline + online RAG · Tool calling
Zero retention by design
Your prompts and source files are not a training dataset.
Military-grade privacy posture
Boundary-first engineering for teams who cannot leak load cases.
Respect for norms & standards
IEC, ISO, Eurocode — cited by clause, not paraphrased from memory.
State-of-the-art verification loop
Retrieve → reason → cite → verify. Not one-shot guessing.
Project management workflow
Indexes follow how projects actually move — not how SaaS wishes they did.
Version control & revision awareness
Content hashing means only what changed gets re-processed.
AI model ecosystem
Provider-verified routing across the models your team already trusts
Vevesh is not locked to one vendor. Our allocator routes engineering RAG, extraction, verification, and massive tool-calling loops across OpenAI, Anthropic Claude, Google Gemini, Groq, Moonshot AI Kimi, DeepSeek, xAI Grok, Alibaba Qwen, MiniMax, GLM, and Cloudflare Workers AI — with metered lanes and provider-native wire formats, not a single-model chat wrapper.
OpenAI
GPT-5.x · GPT-5.4 mini/nano
Structured outputs, vision, and reasoning lanes with provider-native wire formats.
tool calling vision structured JSON
Anthropic
Claude Opus · Sonnet · Haiku · Fable
Anthropic-native tool loops and long-context engineering RAG orchestration.
Claude AI tool use long context
Gemini 3.1 Pro · 2.5 Pro · Flash
Gemini multimodal extraction, summarization, and Cloudflare edge inference paths.
Gemini AI multimodal edge
Groq
GPT-OSS 120B · Whisper STT
Groq-accelerated speech-to-text and ultra-low-latency inference lanes.
Groq inference Whisper fast STT
Moonshot AI
Kimi K2.5 · K2.6 · K2.7 Code
Massive tool-calling range aligned with Moonshot Kimi code and reasoning models.
Kimi AI tool calling code models
DeepSeek
V4 Pro · V4 Flash · R1
Deep reasoning and cost-efficient technical synthesis routes.
DeepSeek RAG reasoning
xAI
Grok 4.x · Grok Build
Grok-family models via direct and Cloudflare Workers AI gateway lanes.
Grok AI gateway
Alibaba Qwen
Qwen3.5 · Qwen3.7 Plus · Qwen3 VL
Qwen vision-language and plus-tier models for multilingual engineering corpora.
Qwen AI VL models
Cloudflare Workers AI
OpenAI · Anthropic · Google · xAI · Qwen
Verified edge gateway ecosystem — route without co-mingling customer KB data.
edge AI gateway Workers AI
Together AI
Moonshot Kimi · Qwen · DeepSeek · MiniMax
Open-model hosting for Moonshot, Qwen, and DeepSeek fallback lanes.
open models hosting
DeepInfra
Kimi · Qwen VL · DeepSeek · GLM · MiniMax
High-throughput open-weight routes with metered billing transparency.
inference API
Fireworks AI
Kimi · DeepSeek · Qwen · GLM · MiniMax
Premium open-model lanes for vision and code-heavy extraction workflows.
Fireworks AI
MiniMax
MiniMax M3
Alternative reasoning lane for long-form report generation.
MiniMax AI
Zhipu GLM
GLM-5.2
GLM-family routes for multilingual standards and annex parsing.
GLM AI
Massive tool calling — Moonshot Kimi & beyond
LangGraph-orchestrated loops retrieve, compute, compare, and format across your indexed corpora. Moonshot AI Kimi K2.x and K2.7 Code lanes are fully aligned with structured tool calling — the same class of multi-step workflows teams expect from frontier code assistants, applied to engineering evidence and standards.
- → Groq-accelerated Whisper speech-to-text for field notes
- → Anthropic & OpenAI structured-output lanes for verification passes
- → Gemini multimodal paths for scanned PDFs and drawing extracts
Works alongside Cursor AI & IDE assistants
Cursor AI, GitHub Copilot, and Claude.ai excel at in-editor code. Vevesh owns what they cannot: persistent folder-native indexes, IEC/ISO citation grounding, team KB RBAC, and evidence packs your lead can sign.
-
Cursor AI
Complements in-IDE coding — Vevesh owns folder-native RAG and audit citations.
-
GitHub Copilot
Code completion inside repos; Vevesh indexes specs, reports, and standards beside them.
-
Microsoft Copilot
M365-embedded Q&A; Vevesh covers mixed engineering folders Copilot does not index natively.
-
Perplexity
Web search answers; Vevesh grounds on your project files with page-level evidence.
-
Claude.ai
General chat; Vevesh adds persistent KB indexes, RBAC, and verification loops.
-
ChatGPT
Session uploads; Vevesh replaces one-off attachments with living folder indexes.
Extraction
Four engines. One knowledge base. You pick the depth.
Local or server extraction per knowledge base — from lightning-fast text pulls to premium vision for the messiest scans.
Fastest
Express
Plain text layer extraction. Ideal for digital PDFs, Word docs, and clean reports where speed matters.
Best for: Daily specs, code, markdown
Layout-aware
Structure
Preserves headings, tables, and document hierarchy. Cloudflare toMarkdown for complex Office and PDF layouts.
Best for: Technical reports, tables
Vision OCR
Flow
Workers AI vision for scanned images and messy documents. Falls back intelligently on unsupported formats.
Best for: Scanned drawings, photos
Highest fidelity
Premium
Higher-quality vision model for the toughest scans — handwritten margins, low-contrast plots, legacy archives.
Best for: Legacy archives, field notes
Local or server — your call
Extract on your machine for full control, or push raw files to the edge pipeline when you want server-side processing.
Local extract Server pipeline
Use cases
Where Vevesh earns its keep on day one
Design partners use Vevesh wherever project knowledge is folder-shaped, revision-heavy, and too valuable to re-upload every Monday.
Wind & renewable energy
Standards libraries that actually answer
Index IEC extracts, foundation reports, and site assessments together. Ask load-combination questions with clause-level citations — not paraphrased guesses.
IEC 61400Load casesSite reports
Civil & structural
Rev-controlled report corpora
Rev A through Rev E live in one folder tree. Vevesh hashes incrementally so only changed documents re-index — critical when submissions land weekly.
EurocodeRev controlPDF + Word
Software & R&D
Code + docs in one query surface
Mix README files, API specs, and architecture PDFs. Hybrid retrieval finds the implementation detail and the design rationale in one grounded answer.
MonoreposAPI specsHybrid RAG
How it works
From messy folders to trustworthy answers in three moves
The workflow mirrors how you already organize work — Vevesh adds the AI layer without forcing a new filing system.
- 01
Pick a folder
Choose any project directory — specs, reports, CAD exports, code. No forced “import everything to our cloud” step.
- 02
Index & extract
Vevesh scans, hashes, and extracts with the engine you select. Local on your machine, or server-side via the edge pipeline.
- 03
Ask in context
Chat with full project awareness — citations, cross-document reasoning, and optional shared knowledge bases for the team.
Platform
Serious infra behind a calm interface
Hybrid stack: local indexing on your machine, managed retrieval where it helps, edge workers for scale-sensitive paths. This marketing site lives on its own Cloudflare account — separate from customer data.
- ✓ Vector + keyword hybrid search
- ✓ Team knowledge bases with Stripe billing
- ✓ Multi-provider AI routing with cost controls
- ✓ Desktop-native filesystem access
Your machine
Desktop app · local index
Edge
CF Workers
Server
Self-host / Dell
Team KB storage
R2 · Vectorize · D1 shards
Consumer AI
ChatGPT
General AI · ephemeral file upload
Engineering fit score 58/100
Pros
- + Instant to try — no setup
- + Broad general knowledge
- + Fast for one-off questions
- + Large model selection
Cons
- − No persistent folder index
- − Upload limits & session-bound files
- − Weak page-level citations
- − Not built for engineering RAG
- − Data leaves your boundary by default
Enterprise suite
Microsoft Copilot
M365-embedded assistant
Engineering fit score 64/100
Pros
- + Deep Microsoft 365 integration
- + Enterprise SSO & admin controls
- + Familiar for Office-heavy teams
- + SharePoint connectivity
Cons
- − Locked to Microsoft ecosystem
- − Limited CAD / engineering tool paths
- − Opaque retrieval across complex folders
- − Per-seat cost at scale
- − Not folder-native for mixed repos
Knowledge wiki
Notion AI
Wiki-first Q&A inside Notion
Engineering fit score 61/100
Pros
- + Clean wiki UX
- + Team pages in one place
- + Simple Q&A over Notion content
- + Good for product/docs teams
Cons
- − Must migrate content into Notion
- − No native PDF/CAD pipeline
- − Engineering folders stay outside
- − No hybrid vector + BM25 stack
- − Citation depth varies by page type
Enterprise search
Glean
Enterprise workplace search
Engineering fit score 72/100
Pros
- + Strong SaaS connector catalog
- + Enterprise sales & support
- + Cross-app search unified UI
- + Mature admin tooling
Cons
- − Six-figure contracts typical
- − Long procurement cycles
- − Not optimized for local project folders
- − Engineering extraction is generic
- − Overkill for technical SMB teams
Engineering AI
Vevesh
Folder-native engineering RAG
Engineering fit score 93/100
Pros
- + Indexes folders you already use
- + 4 extraction engines · local or server
- + Page-level citations on every answer
- + Team KBs with ghost / open modes
- + Hybrid vector + BM25 retrieval
- + Self-host path + isolated CF data plane
- + Desktop-native for heavy engineering repos
Cons
- − Private preview — not self-serve yet
- − Requires design-partner onboarding
- − Best for teams with real technical folders
Design partners
Teams who cannot afford vague AI
"We stopped re-uploading the same IEC PDF every sprint. The folder index just stays current — that alone saved hours."
"Citation paths my reviewers can click beat any summary Copilot gave us on SharePoint-only content."
"Ghost mode for our internal standards KB was the policy knob our legal team actually understood."
Trust & security
Serious teams need more than a shiny chat box
Vevesh is designed for engineers who cannot afford hallucinated load cases or mystery data residency. We optimize for traceability, isolation, and control — not engagement metrics.
Private preview
Design partner program
Self-host
Optional on-prem path
EU-ready
Configurable deployment
Your data, your boundary
Source folders remain on your machine or infrastructure. Index metadata and team KB settings live in Postgres — never scattered across anonymous SaaS uploads.
Isolated data planes
Customer knowledge bases run on dedicated Cloudflare accounts with separate workers, R2, and Vectorize — not co-mingled with marketing or admin traffic.
Transparent citations
Every grounded answer links back to indexed files and pages. Ghost mode hides KB existence from probing; open mode makes citations first-class.
Predictable billing
Team knowledge bases use Stripe subscriptions. Server extraction and embed usage are metered per engine — no surprise token black boxes.
Engineering-grade stack
Hybrid vector + BM25 retrieval, multi-provider AI routing, desktop-native filesystem access, and optional self-hosted server deployment.
Built in the open with partners
Vevesh is in private preview with design partners in wind, civil, and software engineering. We ship against real folder workflows, not demo PDFs.
Integrations & stack
Infrastructure + verified AI provider lanes
Cloudflare edge, Postgres billing, Stripe team KBs — and multi-provider AI routing across OpenAI, Anthropic, Google Gemini, Groq, Moonshot AI, DeepSeek, and xAI Grok.
Desktop app
Native folder access
Cloudflare Workers
Edge extract & index
Vectorize + R2
Team KB storage
Postgres
Settings & billing
Stripe
Team subscriptions
LangGraph
Tool-calling orchestration
Blog
Engineering AI — explained properly
Architecture notes, standards traceability, and pipeline decisions from the team building folder-native RAG.
PrivacySecurity
Zero-retention privacy architecture for engineering AI
A deep technical guide to zero-retention design, military-grade data boundaries, and why engineering teams cannot treat AI like consumer chat — from the architects of Vevesh.
Read article
TeamKnowledge Base
Team knowledge bases — sharing engineering corpora without losing control
Promote a folder into a billed, shared knowledge base with ghost or open citation modes. How Vevesh handles membership, Stripe billing, and retrieval policy.
Read article
ExtractionKnowledge Base
Local vs server extraction — choosing the right engine for your knowledge base
Express, Structure, Flow, or Premium? Local desktop extraction vs server pipeline— a practical guide for engineering document sets.
Read articleFAQ
Questions serious teams ask first
Does Vevesh upload my entire project folder to the cloud?
No. Your source folder stays where it is. Vevesh scans and indexes content locally by default. Team knowledge bases sync extracted chunks and metadata to managed storage — raw files only when you explicitly choose server extraction mode.
How is this different from ChatGPT with file upload?
Upload tools treat files as one-off attachments. Vevesh maintains a persistent, searchable index tied to your folder structure — with hybrid retrieval, engineering RAG routing, and page-level citations on every grounded answer.
Can we self-host?
Yes. Vevesh supports a self-hosted server path for teams that need on-prem control. Managed cloud components (edge workers, team KB storage) are optional layers — not mandatory for every workflow.
What file types are supported?
PDFs, Office documents (Word, Excel, PowerPoint), plain text, code, and common engineering exports. Extraction depth depends on the engine you select — from fast text-layer pulls to premium vision for scanned archives.
How do team knowledge bases work?
An owner promotes a folder into a hosted, billed corpus. Members see it as a sector in chat. Owners control disclosure mode (ghost vs open citations), extraction settings, and Stripe-backed subscription billing.
Which AI models and providers does Vevesh support?
Vevesh routes engineering RAG, extraction, and tool-calling loops across OpenAI GPT, Anthropic Claude, Google Gemini, Groq, Moonshot AI Kimi, DeepSeek, xAI Grok, Alibaba Qwen, MiniMax, GLM, Together AI, DeepInfra, Fireworks AI, and Cloudflare Workers AI gateway lanes. Admins configure verified pricing lanes — not a single locked vendor.
How is Vevesh different from Cursor AI or GitHub Copilot?
Cursor AI and Copilot excel at in-editor code completion. Vevesh complements them with folder-native indexes across PDFs, Office, standards libraries, and project archives — with page-level citations, team KB RBAC, verification loops, and evidence packs for design review.
Does Vevesh use Groq or Moonshot AI Kimi?
Yes. Groq powers accelerated speech-to-text (Whisper) and low-latency inference lanes. Moonshot AI Kimi K2.x and K2.7 Code models support massive tool-calling workflows for multi-step engineering tasks — retrieve, compute, compare, and format with LangGraph orchestration.
Is Vevesh a ChatGPT or Claude.ai replacement?
For general chat, no — use ChatGPT or Claude.ai. For persistent engineering knowledge bases with citation grounding, zero retention boundaries, and multi-provider routing tuned to technical corpora, Vevesh is the purpose-built layer.
Is Vevesh ready for production?
Vevesh is in active private preview with design partners. Core indexing, chat, and team KB flows are live; we are hardening server extraction pipelines and onboarding teams incrementally.
Ready to put your project knowledge to work?
Vevesh is in active development with early design partners. Reach out to join the preview or discuss deployment for your team.
No spam. No sales deck unless you ask. We onboard engineering teams who have real folders to index.