Yeah, me too. And that’s exactly the problem AgentsView solves.

I spent a full afternoon digging into this one — reading the docs, tracing the codebase, and mapping out exactly what it can do. Still, it was worth every minute. Here’s what I found.

TL;DR — What Is AgentsView?

AgentsView (kenn-io/agentsview) is a local-first session intelligence engine for AI coding agents. One Go binary discovers, indexes, and visualizes sessions from 28+ AI coding agents — Claude Code, Codex, Cursor, Copilot CLI, Gemini CLI, OpenCode, Forge, and even Hermes Agent — and stores everything in a local SQLite database.

You get a web UI at http://127.0.0.1:8080 with full-text search, cost dashboards, activity heatmaps, and per-session breakdowns. And here’s the kicker — no accounts, no cloud sync, no sending your data anywhere.

Why AI Session Analytics Matters

Here’s a concrete scenario. You run Claude Code for a 3-hour refactoring session. Then you switch to Codex for a code review. Then Cursor for some quick edits. So at the end of the month, your provider bill arrives and you have no idea which tool cost what.

AgentsView fixes this. So it auto-discovers session directories for every supported agent, reads the raw session files, and indexes them into SQLite. But here’s what makes it different — once indexed, you can:

• Search across all past sessions by keyword (FTS5 full-text)
• Track costs per agent, per model, per project
• Compare agent efficiency — which tool uses more tokens per task
• Analyze usage patterns — when do you code most, which archetypes dominate

I tested the search on the dataset and it’s instant — SQLite FTS5 means sub-100ms queries across thousands of sessions. Meanwhile, tools like ccusage re-parse raw files every time you ask a question. So that’s a massive difference when you’re sitting on 500+ sessions.

AgentsView Core Features — What You Actually Get

Auto-Discovery: Zero Config

AgentsView scans your home directory for known agent session paths. Out of the box it finds 28 agents — everything from the big names (Claude Code, Codex, Cursor) to niche tools (Zencoder, Amp, iFlow, Pi, WorkBuddy).

Notably, Hermes Agent is in the list — which means if you’re running Hermes (like I do), AgentsView indexes those sessions too. That’s a nice touch.

Cost Tracking That Actually Works

The agentsview usage command is the headline feature. It replaces ccusage with something that works across all agents, not just Claude Code.

# Daily cost summary (last 30 days)
agentsview usage daily

# Per-model breakdown
agentsview usage daily --breakdown

# Filter by agent and date
agentsview usage daily --agent claude --since 2026-04-01

# JSON output for scripts
agentsview usage daily --all --json

# One-liner for status bars
agentsview usage statusline

The cost calculation uses LiteLLM pricing with an offline fallback, and it’s prompt-caching-aware — it distinguishes cache creation tokens from cache read tokens. That’s critical for accurate Claude Code cost estimates, by the way.

The agentsview stats command goes further: it classifies sessions into archetypes (quick / standard / deep / marathon / automation), shows tool usage frequency, per-hour activity breakdown, and cache economics. Honestly, the archetype classification alone is worth the download — it tells you whether you’re doing mostly quick-fix sessions or deep architectural work.

Web UI: Keyboard-First, Live Updates

The web UI at http://127.0.0.1:8080 is a Svelte 5 SPA. It’s fast and feels like a native app. Key features:

• Full-text search across all message content (SQLite FTS5)
• Activity heatmap — see your coding patterns over time
• Per-session cost — which models were used, how many tokens
• Live updates via SSE — open sessions stream in real-time
• Keyboard shortcuts — j/k to scroll through sessions, Cmd+K for search, ? for all shortcuts
• Export sessions as HTML or publish to GitHub Gist

Privacy-First

All session data stays local. So the server binds to 127.0.0.1 by default. Still, there’s a minimal anonymous telemetry ping for active users — but it carries no session data, no prompts, no paths. And you can disable it with AGENTSVIEW_TELEMETRY_ENABLED=0 if you want zero outbound.

AgentsView Quick Start: Dashboard in 30 Seconds

Installation is one command:

# macOS / Linux
curl -fsSL https://agentsview.io/install.sh | bash

Then start the server:

agentsview serve

On first run, it discovers all supported agent sessions on your machine, syncs them into the SQLite database at ~/.agentsview/, and opens http://127.0.0.1:8080. From there you get a dashboard with all your sessions indexed and searchable.

I found this particularly slick — no config files, no environment variables, no pointing at directories. It just finds everything. Then you’re in.

AgentsView vs ccusage: Head to Head

The comparison with NVIDIA’s ccusage is the obvious one, since ccusage is the existing tool for Claude Code cost tracking. But the gap is wider than most people expect:

The 100x speed claim isn’t marketing fluff — it’s a direct consequence of the architecture. AgentsView indexes session data into SQLite at ingest time, so queries hit pre-built indexes. But ccusage re-parses raw JSON session files from scratch every time you run it. So if you have 500+ sessions, the difference is seconds vs milliseconds.

Docker Deployment for AgentsView

Most people run AgentsView locally on loopback. But if you want persistent access from multiple machines, or you’re managing a team’s agent costs, the Docker Compose setup is your path.

Here’s the production compose file that ships with the project:

docker compose -f docker-compose.prod.yaml up -d

This deploys AgentsView with a PostgreSQL backend instead of SQLite — which means your session data survives container restarts and is accessible from multiple clients. And you can also push session data from your local SQLite to a central PostgreSQL instance:

docker run --rm -p 127.0.0.1:8080:8080 \
  -e PG_SERVE=1 \
  -e AGENTSVIEW_PG_URL='postgres://user:***@postgres.example.com:5432/agentsview?sslmode=require' \
  ghcr.io/kenn-io/agentsview:latest

The production setup is designed for a VPS. You’d run this on a cheap cloud server, mount your agent session directories, and access the analytics dashboard from anywhere.

So if you’re running multiple agent workflows across your team and need a central cost dashboard, a production AgentsView instance on a DigitalOcean droplet or Vultr VPS is a solid setup. A $12/month VPS handles this easily, and the PostgreSQL backend gives you reliability that SQLite alone can’t match.

Who Should Use AgentsView

Yes, if you:

• Use 2+ coding agents regularly and want to compare costs
• Have API budget anxiety and need per-agent breakdowns
• Manage a team of developers using AI coding tools
• Want a searchable database of past agent sessions
• Already use Claude Code, Codex, Cursor, or any of the 28+ supported agents

Skip it if you:

• Use one agent exclusively and don’t care about analytics
• Never check your API costs
• Prefer a manual spreadsheet approach

The Bottom Line on AgentsView

AgentsView fills a genuine gap in the AI coding tool ecosystem. It’s the kind of tool you didn’t know you needed until you have a $500+ monthly API bill and no idea which agent ran it up.

But the single-binary install, zero-config setup, and broad agent support make it the obvious upgrade from ccusage. And the active development pace (v0.33.1 released two days ago, trending #7 on GitHub) suggests the team is committed to iterating fast.

So here’s my take: if you use more than one coding agent, install AgentsView. It takes 30 seconds, and you’ll learn something about your own coding habits within the first dashboard view.

This is the third piece in our AI coding agent tool chain series — check out the Claude Memory review for agent memory tools, and Agent Skills review for skill frameworks that complement session analytics.