Self-hosted observability for AI coding agents

When FinOps asks about AI coding spend, the first question from security is often: where does the data go?

For many organizations, the answer cannot be “a third-party SaaS that ingests prompts and session logs from every developer laptop.” Agent telemetry includes file paths, error messages, test output, and sometimes full prompts — data that belongs inside your network.

Why self-hosted matters

Self-hosted RetrySight Enterprise runs manager, database, and dashboard inside your infrastructure:

• Data residency — events batch from collectors on developer machines to your manager, not a vendor cloud
• SSO from Core — OIDC and SAML with encrypted IdP config at rest; MCP and ClickHouse unlock at Integration
• Tier gating enforced server-side — capabilities unlock progressively from Core through Cloud Scale
• BYO RDBMS — PostgreSQL by default; MySQL and SQL Server best-effort; Oracle DBA-assisted

RetrySight Cloud exists for teams that prefer managed hosting with multi-tenant isolation and per-account agent keys. RetrySight Lite stays local on loopback for individual developers. The choice is yours.

Deployment in practice

Enterprise Core starts with Docker Compose:

cp .env.example .env
./scripts/build.sh
./scripts/up.sh

Collectors on developer workstations tail local agent logs and push to the manager over authenticated REST ingest. The dashboard queries via GraphQL — no third-party SaaS in the telemetry path.

Higher tiers add cost analytics, MCP, ClickHouse dual-write, and object archive without changing the core architecture.

Who this is for

• Platform teams standardizing on 2+ coding agents across 50+ engineers
• Security teams requiring data residency and SSO
• FinOps partners who need executive PDF exports, not spreadsheet exports from five different vendor consoles

Explore Enterprise tiers or contact us for a security review.