Most of my coding work happens in Claude Code, and from time to time I use Codex or try out other agents. When I remember solving something but not where, I want to search all of that history in one place. The tool I've settled on for that is AgentsView, an open source project from Wes McKinney's company, Kenn Software. I have no relationship with Kenn or with Wes. I found AgentsView after I started using msgvault, another of their tools, and it has become one of the most useful things on my laptop.
It's a narrower kind of record than the one I asked for in my last note. AgentsView reads the transcripts my coding agents already write on my laptop, so it shows what each tool recorded rather than an independent account of everything that happened. For coding agents, that's still the most useful record I have.
AgentsView pulls those session files from Claude Code, Codex, and dozens of other agents into a local archive with a web interface on localhost. Your sessions stay on your machine unless you turn on a feature that shares them. It does send limited usage telemetry by default, with no session content or prompts, and I turned it off with one environment variable.
Finding and resuming sessions
Full-text search covers messages and tool calls across every agent AgentsView supports. There's also an opt-in semantic search that matches by meaning, which sends text to whatever embeddings service you configure, so I've left it off. Claude Code and Codex each have their own history and resume, and tools like ccusage cover costs, but what AgentsView adds for me is one place for all of it.
For Claude Code, Codex, Copilot, Cursor, Gemini CLI, and several others, AgentsView also gives you the command to resume a session in the tool it came from, already pointed at the right project folder. I use it with Claude Code and Codex, and it makes getting from a search result back into the work quick. That only works while the original tool still has the session. Once the tool's own copy is gone, I can still read the transcript in AgentsView, but I can't pick it back up.
The trade-off of keeping history
The archive outlives the tools' own records, and that's a real benefit. By default Claude Code deletes local transcripts after 30 days, a setting you can change, and Anthropic says it does that because transcripts can contain source code and credentials. One user on X likes AgentsView for exactly this reason, since the sessions are still there after the originals are gone.
The same reasoning applies to the archive itself: a longer-lived collection of transcripts is a more valuable target and deserves the same care as any other sensitive data. AgentsView can scan the archive for secrets that ended up in transcripts, which is easy to do when agents read config files and environment variables, and it can prune sessions you no longer want to keep. The scan reports what it finds rather than redacting anything, so the cleanup is still up to you. AgentsView also includes a read-only MCP server that lets an agent search earlier sessions, and I'd only turn that on after scanning, because anything in the archive becomes something an agent can retrieve.
What the cost view shows me
The insight I find most interesting is cost, specifically what my sessions would have cost at API prices. It's an estimate rather than a bill, but I think that number is going to matter more as model providers push heavier users toward paying by the token for their most expensive models. Simon Willison has pointed people on Hacker News to uvx agentsview usage daily for this kind of view, and he wrote that he's been really enjoying AgentsView for tracking token use across his coding agents.
Try it on your laptop
If you use Claude Code, Codex, or another supported agent, you can see your own numbers in about a minute without a full install. You need uv, whose uvx command downloads AgentsView into its cache and runs it, and the first command builds a local archive in ~/.agentsview from the session files already on your machine. Setting AGENTSVIEW_TELEMETRY_ENABLED=0 in your shell first keeps telemetry off. These are the three I'd start with.
Today's estimated cost at API prices, on one line:
uvx agentsview usage statuslineThis week's activity, including active versus idle time and agent-minutes across parallel sessions:
uvx agentsview activity report --preset weekA scan of your transcripts for credentials that leaked into them, with a one-line summary when it finishes:
uvx agentsview secrets scanThen uvx agentsview secrets list shows where each finding came from, with the values redacted. When I ran the scan, it found API keys in my own transcripts. Agents read config files and environment variables all the time, so it's worth running even if you think you've been careful.
What I want next
AgentsView reads sessions from Grok Build, xAI's terminal coding agent, but not from always-on personal agents like Grok Bot, Meta's Muse, or OpenAI's Dots, which run in the vendors' clouds rather than on my laptop. Those are the agents that get the broadest access to email, calendars, and accounts, so they're the ones I most want a record for. Grok Bot is a good example: it can call on other models, and if I don't pay close attention it runs those calls through the API. For now I keep an eye on that myself. I'd like the same searchable, local record for these agents that I now have for my coding agents.
I'm also curious what other people are using. If you keep track of your agents' sessions with a different tool or dashboard, or you've built your own, I'd like to hear about it on X, LinkedIn, or at contact@exagentica.ai.