Skip to main content
Cart

Guides

Power Claude Session Flow Explorer: Mermaid Graphs of Agents & Turns

Neural-LLM Admin 3 min read
Power Claude Session Flow Explorer graph of agents and tool turns
See What ran: prompts, tools, and the Detail inspector — not a four-box HTML mock.

Coding-agent sessions are not flat transcripts. They fan out into sub-agents, tools, gates, and evidence. Power Claude’s Session Flow Explorer maps that structure as a live graph — with Mermaid export for docs — so you can see what actually ran, not guess from JSONL. Try it free in the browser (synthetic demo): Open Flow Explorer — harness-aware demo.

TL;DR

NeedWhere
Try without installingInteractive Flow Explorer demo (start with harness-aware)
Visual session map (product)Power Claude → Flow / Session Explorer → Graph / Combined
Paste into mermaid.live / docsCopy Mermaid or pc tree --format mermaid
Evidence on a tool turnGraph node → Evidence I/O panel
Product deep-linkSession Flow Explorer tile
Product tile: [Session Flow Explorer](/power-claude#feature-flow-explorer) · Free demo: [try/flow-explorer](/power-claude/try/flow-explorer?scenario=harness-aware#scenario=harness-aware) · Related: [Harness Flow Graph](/#harness-flow-graph) (static pipeline map, different surface)

Try it now (no install)

The public demo is the fastest way to understand the product:

  1. Open Try Flow Explorer (recommended scenario: Harness-aware run).
  2. Use Combined mode — session path drawn on the harness map.
  3. Walk Taken path / step nodes; open Evidence on a tool turn.
  4. Switch scenarios (simple edit, failed tool, sub-agent fan-out) from the top dropdown.

Synthetic data only in the browser. Real sessions stay local in the Power Claude extension.

The problem: JSONL is not a map

A long agentic run produces thousands of lines of session log. That is storage, not understanding.

Operators searching for:

  • “claude code session graph”
  • “mermaid agent tree”
  • “what sub-agents did this session spawn”

need a structured view: parent vs child, tool vs model turn, how long each step took, and where evidence lives.

What Session Flow Explorer is

Power Claude’s Flow / Session graph surfaces:

  1. Real sessions from your machine (sandbox mode aligned with production data paths).
  2. Agent fan-out — tasks and sub-agents as first-class nodes, not invent-from-empty ghosts.
  3. Durations — wall time where the log supports it.
  4. Evidence I/O — inspect tool inputs/outputs from the graph.
  5. Mermaid export — same shape for docs and mermaid.live.
  6. CLI parity — pc tree --format mermaid for terminals and CI notes.

Fail-hard validation drops ghost nodes so the map does not invent agents that never ran.

Session Flow vs Harness Flow Graph

SurfaceAnswersWhen to open
Session Flow ExplorerWhat did this session do?Debugging a run, writing a postmortem
Harness Flow GraphWhat will the edit pipeline run for any client?Understanding gates, matchers, sub-agents in the harness
Both are maps. One is **session runtime**. The other is **static harness pipeline** (also on this site as [Harness Flow Graph](/#harness-flow-graph)).

How to use it

In VS Code / code-server

  1. Open Power Claude → Session Explorer (or Flow Explorer entry).
  2. Select a session.
  3. Open Graph (structure) and Lifetime (timeline-oriented view where available).
  4. Click a node for Evidence when you need tool I/O.
  5. Copy Mermaid to paste into docs or mermaid.live.

CLI

pc tree --format mermaid
# or project-specific tree commands documented for your Power Claude version

Why Mermaid

Mermaid is portable: README files, Notion, GitHub, design reviews. Exporting the same graph the UI shows means your documentation is not a hand-drawn fiction of the session.

Frequently asked questions

Is this the same as Claude Code’s built-in history?

No. Claude Code stores transcripts. Flow Explorer projects structure (agents, tools, edges) and exports Mermaid.

Does export send data to Neural-LLM?

No. Graphs and Mermaid source are local. Marketing pages may show example diagrams; your sessions stay on your machine.

Can I deep-link a graph?

Yes — Flow Explorer supports deep links / prefs clamping so a URL opens the intended session context (see product release notes for aliases).

What if the graph is empty?

Empty graph usually means the session log has no indexable structure yet, or validation removed incomplete nodes. Re-run a short agentic task and refresh.

Relation to cross-session messaging?

Messaging coordinates live sessions. Flow Explorer explains what one session did. Use both for fleet ops.

Related

Independent third-party software. Not affiliated with Anthropic.