With the change to token based billing, I pay a lot more attention on what my AI agents are doing. Although some info is available, I was missing the tools to evaluate my token consumption and have an in-depth understanding of what's going on.
Until I discovered AgentsView.
AgentsView is a local-first desktop and web app that reads those files and gives you a proper UI to work with them: browse sessions, search across all message content, see tool calls and thinking blocks, and get a quick health grade for each session. Let's install it and see what it actually looks like.
Installation
There are a few ways in, pick whatever fits your workflow.
As a Windows user I decided to grab the latest .exe or .AppImage from the GitHub Releases page. The desktop app is fully bundled and auto-updates. Executing the downloaded executable will walk you through an installation wizard.
Remark: There is also a CLI which uses the same data directory.
Running it
If you went the CLI route:
agentsview serveThat's it. On first run, AgentsView:
- Creates a SQLite database at
~/.agentsview/sessions.db - Discovers and indexes session files from all supported agents
- Starts watching those directories for new sessions
- Opens the web UI at
http://127.0.0.1:8080
If you went for the Desktop version (like I did), you can just start the installed application:
What you get
Session browser
The main view is a session list across all your projects. You can filter by project, agent, date range, or message count.
Clicking into a session shows the full conversation: user prompts, assistant responses, thinking blocks, and every tool call — the same level of detail as from the Agent tool itself:
This is immediately useful if you've ever lost track of what a long agent session actually did. You can scroll through tool calls, see exactly where it started retrying, and find the moment it decided to restructure your entire folder layout.
Full-text search
All message content is indexed with SQLite FTS5, so you can search across every session. Useful for "what was that conversation where I figured out the right EF Core query" from three weeks ago. Search results link directly to the session and message.
Session intelligence
Each session gets a health score and a session analysis. The score is penalty-based.
Sessions get classified as completed, abandoned, errored, or unknown, each with a confidence level. The classifier uses heuristics — sessions ending on an assistant turn skew completed, sessions ending on a user turn skew abandoned, a final failure streak skews errored. These are signals for triage, not ground truth.
You'll see a grade badge (A through F) in the session header. Clicking it opens a signal panel that shows exactly which penalties were applied.
Token usage and costs
AgentsView includes a Usage view that shows the token count and cost.
 |
Coverage currently includes Claude Code, Codex, Copilot CLI, OpenCode, Gemini, and a few others. The pricing comes from the LiteLLM catalog, fetched on each invocation and cached locally. Pass --offline to use the embedded fallback instead.
Analytics section
The analytics section has activity heatmaps, tool usage breakdowns, velocity metrics, and cost attribution by project, model, or agent. The cost attribution panel includes a treemap and a cache efficiency breakdown, which is handy for spotting prompts that are spending heavily on cache creation without getting many cache reads back.
Multi-agent support
AgentsView supports 29 agents out of the box, including Claude Code, Copilot CLI, Codex, Cursor, Gemini, and a bunch of others I've never heard of. It auto-discovers session directories for the common defaults. If your sessions live somewhere non-standard, point it with environment variables:
There's also a PostgreSQL sync option for multi-machine or team setups, where session data is pushed to a shared database and the UI is served from there.
Should you give it a try?
If you're running AI coding agents regularly and you've ever wanted to understand what a session actually did — not just what files changed, but the full conversation and tool call history — AgentsView is a solid tool for that. The health scoring is a useful triage signal even if the heuristics aren't perfect, and having full-text search across months of session history is genuinely handy.
The cost tracking is a bonus if you're on usage-based billing. Knowing that yesterday cost $23 across three projects before you've had coffee is the kind of feedback loop that changes how you prompt.
It's MIT-licensed, actively developed, and installs in under a minute.
Should you give it a try? In my opinion, absolutely YES!
More information
The docs are thorough if you want to dig into the Session API, PostgreSQL sync, or remote access configuration.