Clausus Documentation

Everything you need to install, configure, and get the most out of Clausus — from first launch to team deployment.

v2.0 · Last updated Feb 2026
⬇️

Installation & Setup

macOS #

Clausus is available as a native macOS application. Requires macOS 13 Ventura or later.

1
Download the installer
Download Clausus-2.0.dmg from clausus.ai/download.
2
Install the application
Open the DMG and drag Clausus to your Applications folder.
3
Download the AI model
On first launch, Clausus will download your chosen model (4–10 GB). This is a one-time download — the model runs locally after this.
4
Start using Clausus
Import your first documents and begin asking questions. No account or cloud connection needed.
Terminal (optional)
# Verify installation
ls /Applications/Clausus.app

# Launch from terminal
open /Applications/Clausus.app

Windows #

Clausus is available as a native Windows application. Requires Windows 10 (21H2) or later, 64-bit.

1
Download the installer
Download Clausus-Setup-2.0.exe from clausus.ai/download.
2
Run the installer
Follow the setup wizard. Default install location is C:\Program Files\Clausus.
3
Download model & start
Same as macOS — select your model on first run, then start working.

First Run #

When you launch Clausus for the first time, you'll be guided through a brief setup process:

  1. Select a model — Clausus automatically detects your hardware and recommends the best model. You can also switch between Swift, Balanced, and Sharp modes.
  2. Set privacy preferences — Configure cloud-assisted processing or run fully offline for air-gapped environments.
  3. Import your first files — Drag documents into the workspace to begin indexing.
💡
Tip
Start with a small batch of 5–10 documents to verify everything works, then import your full library.
🖥️

System Requirements

Clausus adapts to your hardware. AI models run locally, and on lighter machines, encrypted cloud-assisted processing ensures the same quality experience. Here are the specifications.

MinimumRecommended
OSmacOS 13 / Windows 10macOS 14+ / Windows 11
RAM8 GB16 GB or more
Storage10 GB free30 GB+ free
Processor4-core CPUApple M1+ / 8-core CPU
GPUNot requiredMetal (Mac) / CUDA (Win)
Lighter Hardware
8–16 GB RAM
Compact local models + cloud-assisted processing for same quality
Standard Hardware
16–32 GB RAM
Medium local models with moderate cloud assist when needed
Powerful Hardware
32 GB+ RAM
Large local models, minimal cloud usage — mostly on-device
ℹ️
Adaptive AI
Clausus automatically detects your hardware and selects the right model size. Lighter machines get more cloud-assisted processing to match the quality of more powerful setups — all requests are encrypted and ephemeral. Apple Silicon Macs with unified memory offer the best performance per watt.
📖

Lego · Read & Analyze

Lego is the document intelligence module. Import files, index them locally, and ask questions across your entire library with source-cited answers.

Importing Documents #

Clausus supports the following file formats:

FormatExtensionsNotes
PDF.pdfWith OCR for scanned documents
Word.docx, .docFull formatting preserved
Excel.xlsx, .xls, .csvMulti-sheet with table detection
PowerPoint.pptxSlide text and notes extracted
Text.txt, .rtf, .mdPlain and rich text

To import files, either drag and drop into the workspace or click the + Add Files button. You can also import entire folders — Clausus will recursively scan for supported file types.

Each file is automatically parsed, chunked into passages, enriched with LLM-generated metadata (summaries and keywords), and indexed into the local vector store.

⚠️
File size limit
Individual files up to 100 MB are supported. For very large files, processing may take a few minutes on first import.

Chatting with Files #

Once files are indexed, switch to the Chat view to ask questions. Clausus searches across all indexed documents using hybrid retrieval (semantic + keyword) and constructs answers grounded in your data.

Every answer includes source citations — click any citation to view the exact passage and page in the original document.

Tips for effective queries:

  • Be specific: "What was our Q3 revenue?" works better than "Tell me about revenue"
  • Reference specific documents: "According to the audit report, what were the findings?"
  • Ask comparative questions: "How does Q3 compare to Q2 in terms of expenses?"
  • Request summaries: "Summarise the key points of this contract"
✍️

Scribe · Write & Generate

Scribe generates professional documents informed by your imported files. Describe what you need, select the relevant source files, and Scribe produces a draft with inline citations.

Generating Documents #

To generate a document:

  1. Write your prompt — Describe what you want generated. Be specific about format, tone, and length.
  2. Select context files — Choose which indexed documents should inform the output.
  3. Generate — Scribe produces a draft in the preview pane.
  4. Export — Save as DOCX, PDF, or plain text.
Example Prompt
Draft a Q3 executive summary for the board.
Include: revenue highlights, top 3 expense categories,
and strategic recommendations for Q4.
Tone: formal. Length: under 2 pages.
Context: Q3 Financial Report.pdf, Q3 KPI Dashboard.xlsx

Templates #

Scribe ships with built-in templates for common professional document types:

  • Executive Summary — Board-ready overview from financial reports
  • Engagement Letter — Client engagement and scope documentation
  • Compliance Memo — Regulatory compliance summaries for UAE PDPL / KSA
  • Audit Finding — Structured finding with observation, risk, and recommendation
  • Legal Brief — Case analysis with precedent citations

You can also create custom templates by saving any generated document as a template with placeholders.

Miles · Automate

Miles automates repetitive document workflows — extract data from invoices, process batches of files, validate entries against reference data, and export structured results.

Building Workflows #

Workflows consist of steps executed in sequence. Each step performs a specific action on the data flowing through the pipeline.

Available step types:

Step TypeDescriptionExample
ImportScan a folder for filesImport all PDFs from /Invoices/Q3
ExtractPull structured data from documentsExtract vendor, amount, date, VAT
ValidateCross-check against reference dataMatch amounts to bank statement
TransformCompute, format, or restructure dataCalculate VAT totals per vendor
ExportOutput to fileGenerate XLSX with all extracted rows
ConditionBranch based on criteriaFlag if discrepancy > 5%
💡
Tip
Start with one of the pre-built workflow templates (Invoice Extraction, Client Onboarding, Tax Prep) and customise from there.
⚙️

Configuration & Settings

Model Management #

Clausus supports multiple local models. You can switch between them depending on the task — larger models for complex analysis, smaller ones for quick lookups.

ModelSizeRAM RequiredBest For
Phi-3 Medium (14B)8.2 GB16 GB+Deep analysis, complex documents
Llama 3.1 (8B)5.4 GB12 GB+General purpose, fast responses
Qwen 2.5 (7B)4.7 GB8 GB+Arabic/English, multilingual

To switch models, go to Settings → AI Models and click Activate next to your preferred model. Models not yet downloaded will begin downloading automatically.

Privacy Controls #

By default, Clausus operates in fully local mode — no data leaves your device. Available privacy settings:

  • On-device only (default: ON) — Blocks all external network calls for document processing
  • Auto-enrich metadata (default: ON) — Uses local LLM to generate summaries during import
  • Conversation logging (default: ON) — Stores chat history locally for reference
  • Telemetry (default: OFF) — Anonymous usage stats sent to Clausus. Never includes document content.

Cloud API (Optional) #

For tasks that exceed the capability of local models, Clausus supports optional cloud API connections to Anthropic Claude and OpenAI. This is strictly opt-in and disabled by default.

⚠️
Important
Enabling cloud API means selected queries will be sent to external servers. Clausus will always show a clear indicator when a cloud API is being used. You can configure which tasks are allowed to use cloud APIs.
Settings JSON
{
  "cloud_api": {
    "enabled": false,
    "provider": "anthropic",
    "api_key": "sk-ant-...",
    "allowed_tasks": ["complex_analysis", "long_generation"],
    "max_tokens_per_request": 4096
  }
}
🔌

API Reference

Clausus exposes a local REST API on http://localhost:7700 for integration with custom scripts and workflows. The API is only accessible from the local machine by default.

ℹ️
Authentication
The local API uses a bearer token generated during setup. Find it in Settings → API → Token.

Query Endpoint #

POST/api/v1/query
Ask a question across indexed documents. Returns an answer with source citations.
ParameterTypeRequiredDescription
questionstringYesNatural language question
document_idsstring[]NoLimit search to specific documents
top_kintegerNoNumber of chunks to retrieve (default: 8)
streambooleanNoStream response tokens (default: false)
cURL
curl -X POST http://localhost:7700/api/v1/query \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What was our net revenue in Q3?",
    "top_k": 5
  }'
Response
{
  "answer": "Net revenue in Q3 was AED 4.2M...",
  "sources": [
    {
      "document": "Q3 Financial Report.pdf",
      "page": 12,
      "chunk_id": "c_0042",
      "excerpt": "Total net revenue for Q3 2025...",
      "relevance_score": 0.94
    }
  ],
  "model": "phi-3-medium-q4",
  "tokens_used": 842,
  "processing_time_ms": 1240
}

Documents Endpoint #

GET/api/v1/documents
List all indexed documents with metadata.
ParameterTypeRequiredDescription
statusstringNoFilter: indexed, indexing, queued, error
limitintegerNoResults per page (default: 50)
POST/api/v1/documents
Upload and index a new document. Send as multipart form data.
ParameterTypeRequiredDescription
filebinaryYesDocument file (PDF, DOCX, XLSX, etc.)
collectionstringNoAssign to a document collection
DELETE/api/v1/documents/{id}
Remove a document and its indexed data from the local database.

Generate Endpoint #

POST/api/v1/generate
Generate a document from a prompt and context files.
ParameterTypeRequiredDescription
promptstringYesGeneration instructions
document_idsstring[]NoContext documents to inform generation
templatestringNoTemplate ID to use
formatstringNoOutput: markdown, docx, pdf (default: markdown)
streambooleanNoStream response tokens
cURL
curl -X POST http://localhost:7700/api/v1/generate \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Draft a Q3 executive summary for the board",
    "document_ids": ["doc_q3report", "doc_q3kpi"],
    "template": "executive_summary",
    "format": "docx"
  }'

FAQ & Troubleshooting

No. After the initial model download, Clausus runs entirely offline. All document processing, AI inference, and data storage happens on your device. An internet connection is only needed for model updates and the optional cloud API feature.

Your data never leaves your device. Documents, embeddings, conversation history, and all generated content are stored locally in an encrypted database. When you delete a document, all associated data (chunks, embeddings, metadata) is permanently removed.

Clausus grounds all answers in your documents using RAG (Retrieval-Augmented Generation). Every claim includes source citations so you can verify. Local models are slightly less capable than the largest cloud models, but the hybrid retrieval system ensures high precision on document-specific questions. For tasks requiring stronger reasoning, enable the optional cloud API.

Yes. Install Clausus on a shared office machine and connect multiple users from their desks. All team members share the same document index while maintaining separate conversation histories. Admin controls let you manage access levels and lock model configurations.

Response speed depends on your hardware and model choice. To improve performance: switch to a smaller model (Llama 3.1 8B or Qwen 2.5 7B), close memory-intensive applications, ensure you have sufficient free RAM (at least 2x model size), and on Mac, ensure you're running on Apple Silicon natively (not Rosetta).

Clausus is architected to support PDPL compliance by ensuring all data processing occurs locally. No personal data is transferred to external servers. However, compliance also depends on your organisation's broader data handling practices — Clausus is a tool that enables compliance, but your firm should consult legal counsel for a complete compliance assessment.

When online, Clausus checks for updates automatically and notifies you. Updates are downloaded in the background and applied on next restart. For air-gapped environments, download the update package from clausus.ai/download and transfer via USB.

Yes. You can export generated documents as DOCX, PDF, or plain text. Conversation histories can be exported as JSON. The full document index can be backed up from Settings → Storage → Export. Your data is yours — there's no vendor lock-in.