ReductrAI Agent

Open source observability agent. Your data never leaves. Verifiable privacy.

---

Why Open Source?

Every other observability tool asks you to trust them with your data. We ask you to verify us.

• Banks and hospitals need to audit code that processes sensitive logs
• Security teams reject obfuscated binaries they can't read
• Compliance officers want proof, not promises

Our approach: We open-source the agent that touches your data. Read every line. Verify the compression. Confirm nothing leaks.

"We process 100GB of logs to 2KB of math.
 Here's the code that does the shrinking."

---

Quick Start

# Install
curl -sSL https://reductrai.com/install | sh

# Run in local-only mode (FREE, no license required)
reductrai start

# Or with cloud features (requires license)
reductrai start --license YOUR_KEY

Point your telemetry to localhost:8080. Auto-detected format.

---

What's Included

Feature	Description	License Required
Telemetry Ingestion	OTLP, Prometheus, Datadog, and 35+ formats	No (FREE)
Local Storage	DuckDB - query your data with SQL	No (FREE)
Anomaly Detection	Error spikes, latency issues, traffic drops	No (FREE)
Compression Engine	91-95% reduction on telemetry data	No (FREE)
Bring Your Own LLM	Use Ollama, vLLM, or any OpenAI-compatible endpoint	No (FREE)
Cloud Dashboard	Visual UI at app.reductrai.com	Yes
AI Relay	Managed LLM with optimized prompts	Yes

---

Architecture

YOUR INFRASTRUCTURE                        CLOUD (optional)
─────────────────────────────────          ────────────────────
┌────────────────────────────────┐         ┌──────────────────┐
│     ReductrAI Agent (this)     │         │  Cloud Dashboard │
│                                │         │                  │
│  Telemetry In → Compress →     │  ~2KB   │  LLM reasoning   │
│  Store (DuckDB) → Detect       │───────► │  Human approval  │
│                                │summaries│  Remediation UI  │
│  100% of data stays HERE       │         │                  │
│  Code is auditable             │         │  (optional)      │
└────────────────────────────────┘         └──────────────────┘

What stays local:

• 100% of raw logs, traces, metrics
• All PII and secrets
• Full query access (DuckDB)

What leaves (~2KB, optional):

• Pre-aggregated summaries only
• "payment-service: 4.2% errors, P99 523ms"
• No raw data. No PII. Just statistics.

---

Local-Only Mode

Run completely air-gapped with your own LLM:

# Use Ollama
export LLM_ENDPOINT=http://localhost:11434/v1
export LLM_MODEL=llama3.2
reductrai start --local

# Use vLLM
export LLM_ENDPOINT=http://gpu-server:8000/v1
reductrai start --local

What you get:

• Full telemetry ingestion and storage
• Anomaly detection
• Local LLM investigation
• No data leaves your network. Ever.

---

CLI Commands

reductrai start           # Start the agent
reductrai stop            # Stop the agent
reductrai status          # Check status
reductrai query "SQL"     # Run local DuckDB query
reductrai schema          # Show database schema
reductrai version         # Print version

Example Queries

# Error rate by service
reductrai query "SELECT service, COUNT(*) as errors FROM spans WHERE status='ERROR' GROUP BY service"

# Top 10 slowest operations
reductrai query "SELECT service, operation, AVG(duration_us)/1000 as avg_ms FROM spans GROUP BY service, operation ORDER BY avg_ms DESC LIMIT 10"

# Recent error logs
reductrai query "SELECT timestamp, service, message FROM logs WHERE level='ERROR' ORDER BY timestamp DESC LIMIT 20"

---

Building from Source

# Clone
git clone https://github.com/reductrai/agent.git
cd agent

# Build
make build

# Run
./dist/reductrai start

Build Targets

make build        # Build for current platform
make build-all    # Cross-compile all platforms
make release      # Build release binaries
make test         # Run tests
make clean        # Clean build artifacts

---

Configuration

Environment Variable	Default	Description
REDUCTRAI_LICENSE	-	License key (optional for local mode)
REDUCTRAI_PORT	8080	Telemetry ingestion port
REDUCTRAI_DATA_DIR	~/.reductrai	DuckDB storage location
LLM_ENDPOINT	-	OpenAI-compatible LLM endpoint
LLM_MODEL	-	LLM model name

---

Supported Formats

Logs: OTLP, Datadog, Loki, FluentD, Syslog, Splunk, CloudWatch, and more Metrics: Prometheus, Datadog, OTLP, StatsD, InfluxDB, Graphite, and more Traces: OTLP, Jaeger, Zipkin, Datadog APM, Tempo, X-Ray, and more

Point any telemetry to localhost:8080. Auto-detected.

---

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

What We Accept

• Bug fixes
• Performance improvements
• New telemetry format support
• Documentation improvements

---

License

SSPL (Server Side Public License) - See LICENSE

You can:

• Use this for any purpose
• Modify and distribute
• Build commercial products

If you offer this as a service, you must open-source your entire service stack.

---

Support

• Docs: reductrai.com/docs
• Issues: github.com/reductrai/agent/issues
• Email: [email protected]

---

Your data stays local. Code is auditable. AI comes to you.

Get Started | Cloud Dashboard