Skip to content

AI agent workforce analytics

Track all the work your agents are doing in one place

Your agents publish posts, send emails, change ad campaigns, submit apps and deploy code. Each one logs its work to WaveBrain as it goes. You get one searchable record across every agent and machine, stats on how each bot is doing, and failures in your Slack.

1,000 actions a month on the free plan. No credit card required.

The WaveBrain activity timeline with sample data: seven actions from Hermes bots, Claude Code, Codex, Cursor and a script. They include a published Reel, an ad campaign change, support replies, offer codes and a deploy, each with the agent, project, action type, status and time. One App Store submission is marked as failed.

Works with

  • Claude Code
  • Codex
  • Cursor
  • Hermes
  • Windsurf
  • VS Code (Copilot)
  • Claude Desktop
  • Any MCP client or REST call

Questions the timeline answers.

For when agents do your marketing, operations, support and engineering work, often while you are away.

What is the social media bot posting while I sleep?

Open the timeline in the morning. Each post, Story and comment is there with the page, the time and a link.

Did the App Store submission actually go through?

Every entry has a status: success or failure. A submission that failed says so, with what the agent saw when it tried.

Which agent changed the ad budget?

Search for the campaign. The entry names the agent, the project and the time, and the details say what changed.

What went out to customers today?

Filter by email and message. You get every support reply, invoice and newsletter your agents sent, in order.

Which deploy broke checkout?

Filter by project and read the deploys, pushes and database changes from the last hour.

Will I hear about it when something fails?

A failed action posts to your Slack channel with the agent, the project and a link to the entry. You can also send only sensitive actions, such as deploys and payments, or all of them.

Has another agent already done this?

With list_actions, an agent can read the work the others are doing before it starts. A second agent doesn't repeat or undo the first one's work.

MCP tool call
list_actions({
  agent: "codex",
  limit: 10
})

See how your agents are doing, not just what they did.

The Stats page turns the timeline into numbers: how much each bot did, how often it failed and when it last worked. It's how you find the bot that needs attention without reading every entry.

The WaveBrain Stats page with sample data for the last 30 days: 1,284 actions, up 18% on the previous 30 days, a 96.8% success rate, 41 failures and 6 active bots. A daily chart stacks successes and failures, with one day of 9 failures. A table lists each bot with its actions, success rate, failures and when it was last active; the newsletter bot was last active 9 days ago.

Spot a bot that went quiet

Each bot shows when it last did anything. A cron job that stopped or a bot that lost its login stands out days before you notice the missing posts.

Catch a bot that keeps failing

Success rate per bot, and failures stacked in red on the daily chart. A bad day or a bot stuck retrying is visible at a glance.

See if the work is growing

Actions over 7, 30 or 90 days, compared with the period before, broken down by bot, action type and project.

Know which tools they lean on

MCP tool calls by server and tool, with error rates. Names only, never arguments or results, and they don't count as actions.

Your agents' work, posted in Slack.

Pick a channel and WaveBrain posts each action there as it happens: who did it, what changed, and whether it worked. Send everything, only failures, or only sensitive actions like deploys and payments. Free on every plan.

A Slack channel called #agents with sample WaveBrain messages: a Hermes social media bot published a Reel, Claude Code failed to submit an iOS app for review, Codex paused ad keywords, and Cursor deployed a checkout fix. Each message shows the agent, the title, the project, the action type, success or failure, the time, and links to view it in WaveBrain.

Set up in three steps.

WaveBrain is hosted, so there is nothing to run. Your agents reach it over MCP, the same way they use any other tool. Scripts use one REST call.

  1. Connect an agent

    Create an API key, then run the install for the agent you use. Claude Code and Codex install from their plugin marketplaces. Cursor and Hermes have a one-command installer. The setup page has the config for every other agent. A script or bot needs only the API key.

    Terminalone of these
    # Claude Code
    claude plugin marketplace add tambo-ai-labs/wavebrain
    claude plugin install wavebrain@wavebrain
    
    # Codex
    codex plugin marketplace add tambo-ai-labs/wavebrain
    codex plugin add wavebrain@wavebrain
    
    # Cursor, Hermes
    npx wavebrain-cursor install
    npx wavebrain-hermes install
  2. Your agents log as they work

    The MCP server tells each agent to call log_action after anything that changes something outside the chat: a post published, an email sent, an ad campaign edited, an app submitted, a deploy.

    MCP tool callsent by the agent
    log_action({
      agent: "hermes-ops",
      type: "email",
      title: "Sent invoice #1042 to Harbor Supply",
      details: "October retainer, due in 30 days. PDF attached.",
      project: "northwind/billing",
      status: "success"
    })
  3. Read the timeline and the stats

    Filter by agent, project or action type, and search titles and details. Stats shows each bot's volume, success rate and last activity. Connect Slack to hear about failures when they happen.

The agents you already use.

They started as coding tools. People now use them to work in a browser, publish posts, send email, and run ad and app store tasks. WaveBrain logs all of it the same way.

Plugin

MCP server plus a skill that tells the agent when to log, and MCP tool counts for Stats.

  • Claude Code

    plugin

  • Codex

    plugin

  • Cursor

    plugin

  • Hermes

    plugin

MCP server

The agent logs its own actions through the WaveBrain MCP server.

  • Windsurf

    MCP config

  • VS Code (Copilot)

    MCP config

  • Claude Desktop

    MCP config

  • Other MCP clients

    MCP config

REST API

Scripts, cron jobs and bots that don't use MCP log with one REST call and the same API key.

POST /api/v1/actionscurl
curl -X POST https://getwavebrain.com/api/v1/actions \
  -H "Authorization: Bearer $WAVEBRAIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "offers-script",
    "type": "api_call",
    "title": "Generated 500 one-time offer codes",
    "project": "northwind/growth",
    "status": "success"
  }'

What gets logged, and what doesn't.

WaveBrain keeps the actions that reach the outside world. It never asks for your accounts, your code or your conversations.

Recorded

  • What an agent reports

    Anything it logs with log_action, for any kind of work: a post, an email, an ad change, an app submission, a deploy. Each entry has a title, details, the project, a link, a status and optional metadata.

  • MCP tool names, for Stats

    With the Claude Code, Codex, Cursor or Hermes plugin, the server and tool name of each MCP tool call, its status and duration. Never its arguments or results, and it doesn't count as an action.

Not recorded

  • Your accounts

    WaveBrain never signs in to your social pages, inbox, ad accounts or app stores.

  • Your repository and files

    WaveBrain never reads them.

  • Prompts, responses and conversations

    None of it is sent.

  • Reads, searches and research

    They stay out of the activity log, and agents are told not to log them. An MCP read only shows up in Stats as a tool name.

Secrets

Agents are told never to put secrets in an entry they write. WaveBrain doesn't filter that text on the server, so treat an entry like a commit message.

Slack messages carry the title, project, type and status of an action, and leave out its details. The plugins are open source under the MIT license.

Pricing.

Both plans include every agent, the Stats page, Slack alerts and the REST API. Pro adds volume, more API keys and longer history.

Free

$0/month

  • 1,000 actions / month
  • 7-day history
  • Stats for the last 7 days
  • 2 API keys
  • Unlimited Slack channels
  • All agents & clients

Pro

$12/month

  • 50,000 actions / month
  • 90-day history
  • Stats for up to 90 days
  • 50 API keys
  • Unlimited Slack channels
  • All agents & clients

Questions.

Is this only for coding agents?

No. Coding is one kind of work among several. People now use Claude Code, Codex, Cursor and Hermes to work in a browser, publish posts, send email, and run ad and app store tasks. Any action with an effect outside the chat can be logged, whatever kind of work it is.

Does WaveBrain connect to Facebook, the App Store or my ad accounts?

No. WaveBrain has no access to those services. The agent doing the work writes the entry, in plain sentences, right after it acts. That is why it works with any site or tool your agents can use.

Do agents log on their own, or do I have to ask each time?

On their own. When an agent connects, the MCP server gives it instructions to call log_action after any action that changes something outside the chat. You don't need to mention it in your prompts.

What if an agent forgets to log something?

Then that action has no entry. Every connected agent is told to call log_action after anything with an effect outside the chat, and the Claude Code, Codex, Cursor and Hermes plugins add a skill that reminds it when to log. WaveBrain records what agents report.

Is this a tamper-proof audit log?

No. It is a record of what agents report. It works well for reviewing work and finding out what happened. An agent that never reports leaves no entry.

Is this a tracing or observability tool?

No. Tracing tools record every model call inside an app you build. WaveBrain records the outcomes of agents you already use, in plain sentences. It doesn't sit between an agent and its model or tools.

Which agents does it work with?

Any client that supports remote MCP servers. Claude Code, Codex, Cursor and Hermes have a WaveBrain plugin or installer. The setup page has copy-paste config for Windsurf, VS Code and Claude Desktop. Clients that only run local servers connect through mcp-remote. Scripts, cron jobs and bots that don't use MCP log with one REST call.

What happens if WaveBrain is unreachable?

Your agents keep working. A log_action call returns an error to the agent and the task continues.

What data is stored, and where?

Only what your agents send: the title, details, project, link, status and metadata of each action. API keys are stored as hashes. Data is stored in the United States. See the privacy policy for the full list.

What happens when I reach my monthly limit?

New actions are rejected with a quota error until the next month starts or you upgrade. Your existing history isn't affected.

Can I delete my data?

Yes. History older than your plan's window (7 days on Free, 90 on Pro) is purged every day, and you can delete your account and everything in it from the Account page at any time.

Know what your agents are doing today.

Connect your first agent with one install. 1,000 actions a month are free, with no credit card.