Skip to content

How Do I See What My AI Agent Did?

See what your AI agent did: where Claude Code, Codex and Cursor keep history, what git and platform logs show, and how to get one timeline for every agent.

Reviewed by WaveBrain Operations TeamReviewed 6 min read

The short answer

You can see what one AI agent did in its own history, such as Claude Code's local transcripts, Codex's saved sessions or Cursor's checkpoints, plus git and each platform's change log. To see what every agent did in one place, have each agent report its actions to a shared activity log like WaveBrain, which gives you a single searchable timeline and failure alerts in Slack.

You left Claude Code working on a bug fix, a Codex session refactoring tests and a Hermes bot scheduling posts, then went to lunch. An hour later the tests pass, a new Instagram post is live, and a staging deploy you didn't expect has happened. Now you want to know who did what.

The answer depends on whether you're asking about one agent or all of them. Each agent keeps some history of its own, and those are the right place for digging into a single session. For a quick, readable answer across agents and machines, you need them to write to one shared log.

How do I see what my AI agent did?#

You see what one AI agent did by opening that agent's own history, and what all of them did by having each one report to a shared activity log. Built-in histories are per tool and usually per machine, so they work best when you already know which agent and session to check.

Where to lookWhat it showsCovers
Agent's own historyThe conversation, tool calls and file edits of one sessionOne agent on one machine
Git historyCommits, with authors, times and messagesWork that ends up in a repository
Platform change logsChanges inside one account, like Google AdsOne platform account
Telemetry and hooksEvents you export or scripts you run on tool callsOne agent, if you set it up
Shared activity log (WaveBrain)Plain-language actions with status and linksEvery agent and script that reports to it

Where does Claude Code keep a record of what it did?#

Claude Code keeps a full transcript of each session on your machine. Its docs say it stores session transcripts locally in plaintext under ~/.claude/projects/ for 30 days by default, adjustable with the cleanupPeriodDays setting, so you can resume a session or read back what happened.

For more structure, Claude Code can export OpenTelemetry data on usage, costs and tool activity, including tool result events, to a backend you run. Its hooks let you run your own command after a tool call succeeds (PostToolUse) or fails (PostToolUseFailure), which is how you'd write tool calls to a file or service. Our guide on how to log every action Claude Code takes covers those options side by side.

How do I see what Codex did?#

Codex saves your chats locally, so you can reopen them from the terminal. The Codex CLI docs describe codex resume, which reopens a recent chat from the current repository or searches across local chats for older work.

The config reference adds a history.persistence setting (save-all or none) that controls whether Codex saves session transcripts to history.jsonl, and history.max_bytes caps that file by dropping the oldest entries. As with Claude Code, that history lives on the machine where Codex ran.

How do I see what Cursor's agent changed?#

Cursor shows what its agent changed through checkpoints in the chat. Per Cursor's agent docs, the agent automatically creates checkpoints before significant changes, and you can click any checkpoint in the chat timeline to preview your files at that point or restore them.

Restoring reverts files only and doesn't remove messages from the conversation. Checkpoints are great for undoing a wrong turn in the current session, and less suited to answering "what did the agent do last Tuesday?"

What can git and platform logs tell me?#

Git and platform logs tell you what changed inside one system, and they're often the most trustworthy record of that system. The git log command shows the commit logs of a repository, so any work a coding agent committed is visible with times and messages.

Business platforms keep their own histories too. Google Ads change history lists changes to your account, campaigns and ad groups for the past 2 years, including who made a change and previous budget amounts. These records are precise, but each one only knows about its own system, so an email your agent sent or a post it published won't show up in git or Google Ads.

How do I see what all my agents did in one place?#

You see what all your agents did in one place by having each agent log its actions to the same hosted activity log. WaveBrain is a remote MCP server and REST API built for that, and it shows every agent's actions in one timeline you can filter by agent, project and action type, with search over titles and details.

Agents call WaveBrain's log_action tool right after an action with external effects, like publishing a post, sending an email, deploying or changing an ad budget. Each entry has a short past-tense title, details of what changed and why, the project, a status and a link to the result. Multi-step jobs can be grouped into sessions that show one task from start to finish, and failed actions post to Slack with a link to the entry.

Connecting takes one config per agent from the Setup page. Claude Code, Codex, Cursor, Hermes and other MCP clients connect over MCP, and the open-source Claude Code and Codex plugins add hooks that catch high-stakes tool calls like deploys, pushes, payments and emails if the agent forgets to log. Scripts and cron jobs can send a POST to https://getwavebrain.com/api/v1/actions. The agents section of the homepage lists every supported client.

A few honest limits apply. WaveBrain isn't an LLM tracing tool, so it doesn't show prompts, token counts or costs, and it never receives your conversations. It also isn't tamper-proof, since it records what agents report and what hooks catch, and an agent on a client without hooks that never calls log_action leaves no entry. For the transcript of one session, the agent's own history is still the place to look.

Which option should I use?#

Use the agent's own history when you're debugging one session, and a shared activity log when you want to know what every agent did today. Most people who run more than one agent end up using both.

If you only run Claude Code on one laptop, its transcripts and git may be enough. Once you add a second agent, a second machine or a bot that touches customers, a single timeline saves you from opening five tools. For a comparison of logging options, see the best AI agent activity logs for solo founders. To try the cross-agent view, start WaveBrain free. The free plan includes 1,000 actions a month and 7-day history, with no credit card.

Frequently asked questions#

Can I see what my AI agent did after the session closed?#

Yes, if the agent saved its history. Claude Code keeps local transcripts for 30 days by default and Codex can save transcripts to history.jsonl, while a hosted log like WaveBrain keeps entries for your plan's history window.

How do I see what an agent did on another computer?#

Its local history stays on that computer, so you'd need to log in there. A hosted activity log receives actions from every machine, so you can read them from anywhere.

Does WaveBrain read my agent's transcripts?#

No. WaveBrain never reads your repo, files or transcripts and never receives prompts or responses. It stores only what agents and hooks send.

Can I see what my agent did in Gmail or Instagram?#

The platform's own history shows some of it, and the agent can describe the action when it logs to WaveBrain. WaveBrain doesn't sign in to those accounts itself.

What if my agent didn't log an action?#

Check the agent's own history, git and the platform's change log. To close the gap next time, add the WaveBrain logging rules to the agent's instructions or install the Claude Code or Codex plugin, whose hooks catch high-stakes tool calls.

Sources#