DocsSelf-Hosted Evaluation

Self-Hosted Evaluation

Guide

Deploy RADAR in your infrastructure, connect it beside your existing agent stack, and validate evidence collection — all self-hosted, no data leaves your boundary.

Prerequisites#

Infrastructure

  • • Docker Engine 24+
  • • 8GB RAM minimum
  • • 50GB available storage
  • • Linux x86_64 or ARM64

Integration Target

  • • LLM gateway access (logs or stream)
  • • Agent tool activity feed
  • • Optional: SIEM or data warehouse
  • • Network: outbound to your stack

What RADAR does not require

RADAR does not require modifying your agent code, replacing your workflow engine, or routing traffic through a cloud service. It deploys beside your existing stack.

1. Deploy RADAR#

1

PULL THE IMAGE

RADAR ships as a single Docker image. Pull it to your evaluation environment.

bash
docker pull akioudai/radar:latest
2

START THE CONTAINER

Run RADAR with your evaluation license. Data is retained locally.

bash
docker run -d \
  --name radar \
  -p 8080:8080 \
  -v radar-data:/data \
  -e RADAR_LICENSE_KEY=eval-xxxx \
  akioudai/radar:latest
3

VERIFY THE DEPLOYMENT

Confirm RADAR is running and accepting evidence connections.

bash
curl http://localhost:8080/health
# Expected: {"status":"ok","version":"1.0.0","uptime":"12s"}

2. Connect to Your Stack#

1

CONFIGURE EVIDENCE SOURCES

Point RADAR at your LLM gateway logs, tool activity streams, or agent telemetry.

bash
# Configure a gateway log source
docker exec radar radar sources add \
  --type gateway \
  --endpoint https://your-gateway.internal \
  --format openai

Supported sources

LLM gateways (OpenAI, Anthropic, custom), tool execution logs, model responses, SIEM event streams, and webhook integrations.

2

VERIFY EVIDENCE COLLECTION

RADAR begins recording traces, policy findings, and PII signals immediately.

bash
# Check evidence status
docker exec radar radar status

# Expected output
# Sources: 1 connected
# Traces: 247 recorded
# Findings: 12 (2 PII, 3 policy, 7 review)
# Retention: 30 days

3. Review Evidence#

Evidence PipelineObserve \u2192 Detect \u2192 Map \u2192 Export
ObserveLLM calls \u2022 Tool events
DetectPII \u2022 Policy violations
MapAI Act \u2022 GDPR \u2022 SOC 2
ExportPDF \u2022 SIEM \u2022 Evidence packs

Open the RADAR dashboard at http://localhost:8080 to inspect evidence records:

Traces

Reconstruct agent sessions: prompts, tool calls, LLM responses, and decision points in chronological order.

Findings

Review PII exposure, policy violations, missing approvals, cost anomalies, and unresolved exceptions.

Export bundles

Package selected evidence into regulator-ready export files with control mappings and retention metadata.

4. Export Evidence#

1

SELECT EVIDENCE RANGE

bash
# Export last 7 days of evidence
docker exec radar radar export \
  --from "2026-04-27" \
  --to "2026-05-04" \
  --format json \
  --output /data/export-2026-05-04.json
2

GENERATE EVIDENCE PACK

Evidence packs include traces, findings, control mappings, and retention attestations.

bash
# Generate a complete evidence pack
docker exec radar radar pack \
  --id export-2026-05-04 \
  --include-controls \
  --include-retention-proof
# Output: /data/pack-export-2026-05-04.zip
3

SEND TO SIEM

Configure RADAR to forward evidence events to your SIEM for centralized monitoring.

Next Steps#

Architecture

Understand the system boundary and design principles

Evidence Model

Learn about trace records, findings, and control mapping

Previous← QuickstartNextArchitecture →