OwlSight

Appearance

Getting Started ​

Set up OwlSight and run your first AI code review in under five minutes.

Prerequisites ​

  • .NET 10 SDK or later
  • Git installed and available on PATH
  • Access to an OpenAI-compatible LLM API (OpenAI, Ollama, Azure OpenAI, etc.)

Install ​

From Source ​

bash
git clone https://github.com/radaiko/OwlSight.git
cd OwlSight
dotnet build

The CLI binary is at src/OwlSight.Cli/bin/Debug/net10.0/OwlSight.Cli.dll. Run with:

bash
dotnet run --project src/OwlSight.Cli -- <command>

Docker ​

bash
docker build -t owlsight .

See Docker guide for usage.

Initialize a Project ​

Navigate to your git repository and run:

bash
owlsight init
Initialized .owlsight/ directory
  Config:  .owlsight/config.json
  Rules:   .owlsight/rules/

Edit .owlsight/config.json to customize your settings.
Add review rules as markdown files in .owlsight/rules/.

This creates:

.owlsight/
├── config.json                 # Review configuration
└── rules/
    └── no-console-writeline.md # Example review rule

Run Your First Review ​

Make some changes on a branch, then run:

bash
owlsight review --base main --api-key $OPENAI_API_KEY

OwlSight will:

  1. Diff your current branch against main
  2. Load custom rules from .owlsight/rules/
  3. Send the diff to the LLM with tool-calling capabilities
  4. Review — the LLM analyzes changes, uses tools to read surrounding code, and produces findings
  5. Output findings to the console with severity, file:line, description, and suggestions

Example Output ​

  src/Services/AuthService.cs
  CRITICAL  Hardcoded API key (src/Services/AuthService.cs:15)
    API key is hardcoded in source code.
    Suggestion: Use environment variables or a secrets manager.

  WARNING   Missing input validation (src/Controllers/UserController.cs:42)
    User input passed directly to service without validation.
    Suggestion: Add input validation before processing.

  ╭──────────────────────────────╮
  │ Review Summary               │
  ├──────────────┬───────────────┤
  │ Files        │ 3             │
  │ Findings     │ 2             │
  │ Critical     │ 1             │
  │ Warning      │ 1             │
  ╰──────────────┴───────────────╯

  Review FAILED — critical issues found.

Using a Local LLM ​

OwlSight works with any OpenAI-compatible API. To use Ollama:

bash
# Start Ollama with a model that supports tool calling
ollama serve

# Run review against the local endpoint
owlsight review --base main \
  --base-url http://localhost:11434/v1 \
  --model llama3 \
  --api-key ollama

Saving a JSON Report ​

Add --output to write a machine-readable JSON report:

bash
owlsight review --base main --api-key $KEY --output report.json

The JSON report includes version, timestamp, summary counts, and the full findings array. See JSON Output for the schema.

Next Steps ​