No agents to wrap, no servers to run
Dendro discovers existing session transcripts from Claude Code, Codex, Gemini/AGY, Grok, Pi, and OpenRig directly on disk. It reads only new lines as sessions grow.
How Dendro works
From scattered local logs to clear operational answers.
Every AI coding tool leaves behind local session files on your machine. Dendro reads those files incrementally, organizes them into a fast local analytical store, and lets you explore them visually or in plain English.
Separate work, waiting, and rework
Each session is mapped into five clear time bands—active model/tool execution, waiting on a person, waiting on a peer handoff, stalled with work waiting, or parked—plus token and cache cost.
Terminal cockpit, browser report, or plain English
Browse live in the interactive terminal cockpit, ask questions in plain English, or export a privacy-scrubbed HTML report that opens offline in any browser.
Interactive Scenario Model
Estimate the hidden cost of multi-agent rework.
When tasks bounce between roles or stall after context resets, token spend and engineering hours compound fast. Adjust the sliders below to model your team's workflow.
Returned missions / week10.0
Reclaimable tokens / week1.58M
Weekly token spend saved$9.45
Engineering hours reclaimed5.25 hrs
Effective capacity gain+8.8%
This calculator models your assumptions above—it is not a benchmark guarantee. Run Dendro on your own logs to measure your actual rework rate.
Honest boundaries
What Dendro proves—and what it refuses to guess.
Trustworthy engineering telemetry starts by naming what the data can and cannot support.
| Question | What Dendro reports from local evidence | What Dendro never fabricates |
|---|---|---|
| Where did the time go? | Exact wall-clock breakdown across active execution, waiting on humans, waiting on peer handoffs, stalled queues, and parked time. | Subjective "developer focus" scores or surveillance keystroke tracking. |
| Did the workflow converge? | First-pass completions versus multi-pass return loops between seats, plus median handoff wait and execution durations. | Claims about whether the underlying product strategy or architecture was right. |
| Why did token costs spike? | Model-level pricing, cache read/write ratios, and context-window drops pinpointed on the session timeline. | Speculative "ROI multiplier" claims or invented "hours saved vs. a human". |
| Can I share a report safely? | Offline HTML viewer with automatic seat aliasing (Seat 1, Seat 2) and scrubbing of paths, prompts, and tokens. |
Silent background uploads to any hosted telemetry cloud. |