Skip to content

Feature request: show timestamps and per-agent token usage in the chat UI #9

@stefanpieter

Description

@stefanpieter

Feature request

Please add better visibility for timestamps and per-agent token usage in the ACP chat UI.

What I want to see

1. Timestamps

For chat activity, I want timestamps visible so I can tell:

  • when a message arrived
  • when a tool started/finished
  • when an agent handoff happened
  • how long the system was quiet between updates

This is especially useful when diagnosing whether an agent is still working, stalled, or has gone quiet too early.

2. Token usage per agent

In multi-agent / delegated workflows, I want to see token usage broken down by agent or role.

Examples:

  • Lead: input/output tokens
  • Reviewer: input/output tokens
  • Validator: input/output tokens
  • Sub-agent X: input/output tokens

Even a lightweight per-agent breakdown would be very helpful for understanding:

  • which agent is doing the heavy lifting
  • where cost is going
  • whether delegation is actually worth it

Why this matters

For orchestration-heavy workflows, the current UI can feel too opaque. Timestamps and token usage make it easier to:

  • debug workflow silence or latency
  • understand handoffs between agents
  • estimate cost/efficiency by role
  • trust what the system is doing over time

Suggested UI options

Possible implementations could include:

  • timestamps next to messages / tool events / handoff events
  • hover or expandable metadata per message/turn
  • per-agent token badges or a summary panel
  • a compact analytics/timeline view for the active session

Scope

This request is specifically about visibility in the ACP Client UI for:

  • event timing
  • per-agent token usage

not just aggregate session totals.

Additional notes from the original thread

Comment 1

The per-agent token visibility is critical for multi-agent setups — without it you're flying blind on costs.

A few things that help in the meantime:

  1. Claude Code Memory Kit (free on ClawHub) includes a token tracking pattern — it uses a lightweight session manifest that records model/tokens per tool call, making it easy to see which agent is burning the most.

  2. For timestamp visibility, wrapping tool calls in a context manager that logs start_time, end_time, and agent_id gives you a clean audit trail you can diff later.

The deeper issue is that multi-agent UIs are still treating agents as monolithic chat threads rather than concurrent processes with separate budgets. Your request is pointing at exactly the right abstraction.

+1 from our team — this would make debugging agent fleet behavior 10x easier.


Originally filed by mistake in formulahendry/vscode-acp: formulahendry/vscode-acp#25

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions