supavision 0.1.0

AI-powered infrastructure monitoring that discovers and watches your servers intelligently

Supavision

Point it at a server, and it figures out the rest.

Supavision uses Claude to explore your infrastructure, understand what's running, and monitor it — without you writing check scripts or defining metrics. It also scans codebases for security issues and uses AI agents to evaluate and fix them.

Why Supavision?

Traditional monitoring requires you to define every check, threshold, and alert rule upfront. Supavision flips this:

• Discovery, not configuration. Point it at a server via SSH. Claude explores the system, finds running services, databases, and configs, and builds a baseline of what "normal" looks like.
• Drift detection, not threshold alerts. Instead of "CPU > 90%", Supavision detects "this service wasn't running yesterday" or "the config file changed since last check."
• Zero LLM cost. Uses Claude Code CLI (included with your Claude subscription). No per-token API charges.
• Codebase scanning included. 81 security patterns across 9 languages, with AI-powered evaluation that separates real issues from false positives.

Two Capabilities

Infrastructure Monitoring

Add server → Discovery (Claude explores via SSH) → Baseline → Scheduled health checks → Slack alerts

Supports: Servers (SSH), AWS Accounts, Databases, GitHub Orgs — extensible via templates.

Codebase Scanning

Add codebase → Scan (81 regex patterns) → Evaluate (AI agent) → Approve → Implement (AI creates fix)

Findings go through a lifecycle: Scanned → Evaluated → Approved → Implementing → Completed. False positives are learned and auto-dismissed in future scans.

Quick Start

# Install
git clone https://github.com/devsquall/supavision.git
cd supervisor
python -m venv .venv && source .venv/bin/activate
pip install -e .

# Verify setup
supavision doctor

# See it with sample data
supavision seed-demo
supavision serve --port 8080

Open http://localhost:8080 — you'll see a dashboard with sample resources, health history, and code findings.

Prerequisites: Python 3.12+ and Claude Code CLI (for infrastructure monitoring). Codebase scanning works without Claude CLI.

Features

Feature	Details
Web dashboard	Real-time status, resource management, live agent output (SSE streaming)
6 resource types	Server, AWS Account, Database, GitHub Org, Codebase, custom
Codebase scanner	81 regex patterns across Python, JS/TS, Go, Rust, Java, C/C++, PHP, Ruby
AI evaluation	Claude analyzes findings for real exploitability, not just pattern matches
Automated fixes	Approved findings can be fixed by an AI agent in a git branch
Blocklist learning	False positives are remembered and auto-dismissed in future scans
REST API	Full API with key auth and OpenAPI docs at /docs
MCP server	9 tools for querying resources, reports, findings, and blocklist from Claude CLI
Slack alerts	Smart dedup (same issue re-alerts once per day), SSRF-protected webhooks
Scheduled monitoring	Cron-based discovery + health checks with jitter and concurrency limits

CLI

Infrastructure

supavision resource-add prod-web --type server \
  --config ssh_host=10.0.1.5 ssh_user=ubuntu
supavision run-discovery <resource_id>
supavision run-health-check <resource_id>
supavision set-schedule <resource_id> --health-check "0 */6 * * *"
supavision notify-configure <resource_id> --slack-webhook https://hooks.slack.com/...

Codebase

supavision resource-add my-app --type codebase --config path=/home/user/myapp
supavision scan <resource_id>
supavision findings <resource_id>
supavision evaluate <work_item_id>
supavision approve <work_item_id>
supavision implement <work_item_id>
supavision scout <resource_id> --focus security

Operations

supavision serve --port 8080       # Web dashboard + API
supavision run-scheduler           # Cron-based scheduling
supavision doctor                  # Health check
supavision seed-demo               # Sample data for evaluation
supavision purge --days 90         # Cleanup old data

Docker

docker compose up -d

Dashboard at http://localhost:8080. Data persists in the supervisor-data volume.

Configuration

Variable	Default	Description
SUPAVISION_PASSWORD	(none)	Dashboard login password (set this in production)
SUPAVISION_BACKEND	claude_cli	Backend: claude_cli (free) or openrouter (API)
OPENROUTER_API_KEY	(none)	Required if backend is openrouter
SLACK_WEBHOOK	(none)	Global fallback Slack webhook URL
SUPAVISION_MODEL	anthropic/claude-sonnet-4	Model for investigation
SUPAVISION_CHECK_INTERVAL	60	Scheduler check interval (seconds)
SUPAVISION_CLI_TIMEOUT	900	Claude CLI timeout (seconds)
WEBHOOK_ALLOWED_DOMAINS	(none)	Comma-separated webhook domain allowlist

MCP Server

Supavision includes an MCP server that lets Claude CLI query your monitoring data in conversations.

supavision mcp-config  # Print config for Claude CLI

9 tools available:

Tool	Description
supavision_list_resources	All resources with current severity
supavision_get_latest_report	Latest health check report
supavision_get_baseline	Discovery baseline + checklist
supavision_get_run_history	Recent runs with status
supavision_list_findings	Codebase findings with filters
supavision_get_finding	Full finding details
supavision_get_project_stats	Finding counts by stage
supavision_list_blocklist	Known false-positive patterns
supavision_search_findings	Search across all findings

Architecture

                          Resource
                         /        \
              Lane 1: Health       Lane 2: Work
              (infrastructure)     (codebase)
                   |                    |
         Engine → Report →       Scanner → Findings →
         Evaluation → Alert      Evaluate → Implement

Two parallel data pipelines sharing a common Resource model. Infrastructure monitoring produces aggregate health reports. Codebase scanning produces per-issue findings with individual lifecycles. See ARCHITECTURE.md for details.

Tech stack: Python 3.12+, FastAPI, HTMX, SQLite (WAL), Claude Code CLI.

Adding Resource Types

Create templates/{type_name}/discovery.md and health_check.md with {{placeholder}} syntax, then add an entry to resource_types.py. See ARCHITECTURE.md for template placeholders.

Security

Supavision runs AI agents on your infrastructure. Read SECURITY.md for the full threat model, tool scoping, and deployment recommendations.

Key points:

• Credentials stored as env var references, never the actual secrets
• Infrastructure agents use allowlisted read-only commands
• Codebase implementation agents require explicit user approval
• SSRF-protected webhook dispatch
• Dashboard auth via SUPAVISION_PASSWORD

Contributing

See CONTRIBUTING.md for setup, testing, and the two-lane architecture rules.

pip install -e ".[dev]"
pytest tests/ -v          # 528+ tests
ruff check src/ tests/    # Linting

License

MIT