cinsights 0.1.1

Coding agent insights for teams

cinsights

Coding agent insights for teams

AI coding agents are transforming how teams build software. But when your team uses Claude Code, Cursor, and Codex across dozens of projects, you have no visibility into how they're being used, where the friction is, or whether things are getting better or worse over time.

cinsights helps engineering teams track, understand, and improve how their developers work with AI coding agents. Not per-session logs, but patterns across time, across agents, and across your whole team. Which projects have the most friction? What CLAUDE.md rules would help everyone? How is each developer's agent effectiveness changing over time? What patterns are your best developers using that the rest of the team could adopt?

Per-project digests - what's working, what's hindering, quick wins, and ambitious ideas. Aggregated across sessions over days or weeks, not a single-run snapshot.

Per-developer profiles - work areas, interaction style, tool preferences, and how each developer uses coding agents. Built from cross-session patterns, not self-reported surveys.

Grounded friction analysis - recurring pain points linked to specific sessions with impact estimates. Not "you had errors" but "you have a repeated read-before-edit pattern that costs ~40 tool calls per session."

Actionable fixes - copy-paste CLAUDE.md rules and feature recommendations (hooks, custom skills, plan mode) generated from your team's actual friction patterns. Each suggestion is grounded in session evidence.

Quick start

# Install
pip install cinsights
# or: uvx cinsights

# Configure LLM (interactive)
cinsights setup
# No API key? Use Ollama instead:
# ollama pull qwen2.5:14b
# cinsights setup --provider openai --model qwen2.5:14b --base-url http://localhost:11434/v1

# Index + analyze local Claude Code / Codex sessions
cinsights refresh --source local --hours 8760

# Generate a project digest
cinsights digest project my-project --days 30

# Start the web UI
cinsights serve

Open http://localhost:8100. See the getting started guide for the full walkthrough.

Data sources

Source	What it reads	Best for
Local	~/.claude and ~/.codex session files	Try in 2 minutes. No external dependencies.
Entire.io	Git-based checkpoints across Claude Code, Cursor, Codex	Cross-agent and cross-machine coverage for teams.
Phoenix	Arize Phoenix traces	Centralized team observability.

Documentation

• Getting started - install, configure, first run
• Concepts - pipeline, quality metrics, scoring, insights, digests
• Configuration - env vars, config file, CLI reference
• Sources: Local · Entire.io · Phoenix
• Self-hosting - run cinsights on your infrastructure