Coding Agent FinOps & Retry Observability

Your agents retry more than you think.

Fleet observability for AI coding agents — retries, token waste, and session timelines across Cursor, Claude Code, Codex, and more.

Download RetrySight Lite — free Start Cloud trial Explore Enterprise tiers

RetrySight Lite — free, open source, on GitHub · RetrySight Cloud — managed SaaS · 4 self-hosted Enterprise tiers · Watch the Cloud tour

RetrySight Cloud Live view showing task and collector activity with a waterfall of session spans

Collectors for the tools your fleet already uses

CursorClaude CodeCodexKiroWarpCopilotClaude Cowork

See RetrySight Cloud in action

Short narrated videos, one for every page of the portal. Start with the dashboard.

Watch all 16 videos

Choose how you run RetrySight

Lite for solo devs, Cloud for growing teams, Enterprise for platform orgs.

Open source · Popular

RetrySight Lite

Free

Apache 2.0

Local-first desktop app for individual developers. In-process collectors, no Docker, no fleet manager.

Flutter desktop app (macOS, Windows, Linux)
In-process collectors — Cursor, Claude Code, Warp, and more
Loopback API, GraphQL, and MCP
No separate agent service or Docker required

Download

SaaS

RetrySight Cloud

Starter

managed

Managed fleet observability for small and medium teams. We host the manager and dashboard; agents run on developer laptops.

Managed SaaS portal — early access
Per-account agent API keys (rk_live_*)
Multi-tenant isolation by organization
Same retry taxonomy and cost views as Enterprise

Open Cloud

Self-hosted

RetrySight Enterprise

4 tiers

your infrastructure

Full fleet control plane on your network. Four installation tiers from core ingest to cloud-scale archive and push ingest.

Docker deploy with BYO PostgreSQL (MySQL/SQL Server best-effort)
MCP, ClickHouse, and OTel at higher tiers
Tier gating enforced server-side
Data residency — no third-party SaaS in the telemetry path

Explore tiers

Self-hosted in four tiers

Enterprise runs on your network — from free Core ingest to Cloud Scale push ingest and object archive.

See Enterprise tiers & pricing How it works

Numbers platform teams ask for

RetrySight answers the FinOps questions vendor dashboards skip.

Self-hosted Enterprise tiers

Supported coding agent sources

<15m

Target time to first agent heartbeat

From the blog

FinOps, retry waste, and fleet visibility for teams running multiple AI coding agents.

October 1, 2026

RetrySight Lite 1.1.0: accurate retry counts for Claude Code

A release that fixes badly inflated retry_count and token-usage numbers for real Claude Code CLI sessions, plus a backfill tool for data ingested under the old logic.

Read article

September 15, 2026

Lite, Cloud, or Enterprise: a field guide to running RetrySight

Same retry taxonomy, three deployment models. A practical decision framework for solo developers, growing teams, and platform orgs choosing where their agent telemetry lives.