AgentPM
Sign inRequest access
BETA ONBOARDING

Get your first agent sessions into AgentPM.

Install the local collector, confirm your org is receiving data, then turn coding-agent transcripts into searchable work history, key decisions, and coaching signals.

Open AgentPM Back to home
agentpm.dev/orgs/your-org/overview
AgentPM overview with activity heatmap and coaching categories.

// five steps to useful data

Five steps to useful data

  1. 01

    Sign in with your beta account

    Use the email, GitHub, or Google account tied to your AgentPM invite. After sign-in, open the organization you want to connect.

    The beta is invite-gated so captured sessions stay attached to the right team from the first upload.

    AgentPM sign-in page with GitHub, Google, and email options.
  2. 02

    Click Add data in your org

    The Add data dialog gives you the recommended one-line install command for the current organization.

    Copy the direct install command when you want every captured Claude Code, Codex, Cursor, or Grok session routed to this org automatically.

    $curl -fsSL 'https://agentpm.dev/install.sh?org=your-org&code=...' | bashcopy
    Add data dialog showing the direct install command for an organization.
  3. 03

    Run the installer locally

    Paste the command into the machine where agent work happens. For most beta users, monitoring one home directory is the right first choice.

    AgentPM watches local Claude Code, Codex, Cursor, and Grok session files, then uploads normalized chunks without changing how your agents run.

    Terminal showing the AgentPM installer completing on an Ubuntu machine.
  4. 04

    Verify the agent is sending

    Open Admin to confirm heartbeats, upload volume, versions, mapped directories, and machine aliases.

    Healthy agents show recent heartbeat and last-sent timestamps. Admin is also where you clean up naming before rolling out to teammates.

    AgentPM Administration page listing connected org agents and upload statistics.
  5. 05

    Use Overview to find the work

    Once sessions arrive, Overview shows activity by user or directory, searchable conversations, and coaching patterns across the org.

    Start here when you want to know which projects are active, where agent work is clustering, and which patterns deserve PM attention.

    Go to organizations
    AgentPM Overview page with activity heatmap and coaching summaries.

// read the work

Read the work without losing the evidence

AgentPM is most useful once the raw session, the summarized PM signal, and the coaching view sit next to each other. Use these views together when you review a complex agent-assisted task.

Replay the actual conversation

Notebook view keeps the human request, agent replies, command output, and tool activity together so PMs and leads can understand what happened without asking the developer to reconstruct it.

Conversation notebook showing user, coding agent, and command execution blocks with a minimap.

Extract keyframes and decisions

Insights turns long sessions into dated keyframes: decisions, artifacts, state changes, incidents, and high-impact implementation moments with references back to turns.

Conversation insights page with AI-generated keyframes and references.

Coach agent usage from evidence

Coaching flags planning misses, skill gaps, and repeatable habits from the transcript itself, then points back to exact turns for review.

Conversation coaching page showing a planning finding and transcript references.

Start capturing agent work.

One org, one machine, one command. Sessions start landing in Overview within minutes.

$curl -fsSL 'https://agentpm.dev/install.sh?org=…' | bashcopy
Open AgentPM Back to home