How do I monitor LangChain Deep Agents?

Install ClawMetry (pip install clawmetry). The daemon auto-detects ~/.deepagents/.state (the LangGraph checkpointer's msgpack-encoded state) and renders every run, tool call, and per-node token in one dashboard at http://localhost:8900. Decoded read-only.

Does ClawMetry show the full nested sub-agent tree?

Yes. Deep Agents nests LangGraph agents multiple levels deep; ClawMetry walks the checkpointer state and renders the full tree with per-node token totals rolled up under each parent. A parent that spawns 4 nested sub-agents that each burn 50k tokens shows the 200k total under the parent, not four separate mysteries.

Which Deep Agents providers does ClawMetry cover?

All of them. Deep Agents can route to Anthropic, OpenAI, Groq, local models via Ollama, or any LangChain-compatible provider; ClawMetry parses the usage_metadata block on every turn regardless of provider and normalises to input/output/cache/reasoning.

Is ClawMetry safe to point at a running Deep Agents install?

Yes. The checkpointer state file is opened read-only; the msgpack decoder is vendored in the clawmetry-pro wheel (no runtime dependency on LangGraph beyond the state format) so a Deep Agents upgrade cannot break the read.

How much does ClawMetry cost for Deep Agents?

Free forever on your own machine. Cloud starts at $9/node/mo (Starter) or $19/node/mo (Pro).

Observability and cost tracking for LangChain Deep Agents

See every Deep Agents run, tool call, and token as it happens. ClawMetry reads the checkpointer state Deep Agents writes to ~/.deepagents/.state and renders runs, tool trees, tokens, and per-provider cost in one dashboard.

Supported runtime Reads LangGraph checkpointer state (msgpack-encoded), decoded read-only Read-only Free trial

What ClawMetry sees on Deep Agents

ClawMetry's Deep Agents adapter opens the store Deep Agents already writes (~/.deepagents/.state (and ~/.deepagents for the workspace)) in read-only mode and surfaces:

Sessions: started at, last activity, model, message count, per-session token totals.
Live activity: every tool call as it lands, with args and result. Turn-level tokens split.
Per-turn tokens: Per-turn usage extracted from LangGraph checkpoints (LangChain's usage_metadata shape: input_tokens, output_tokens, plus cached / reasoning where present).
Cost per session, per model, per hour: cache-aware where the model supports caching.
Sub-agent-of-sub-agent trees (Deep Agents nests LangGraph agents multiple levels deep): the Deep Agents-specific signal ClawMetry surfaces that generic dashboards do not.
Approval and kill switch: watch for token-maxing signals (message repeat, no-progress heuristic, cost-per-hour threshold), pause or kill before the next call.

Source: the adapter is at clawmetry_pro/adapters/deep_agents.py. Every parse rule and every SQL SELECT is in that file.

Enable it in about two minutes

1. Install ClawMetry

Pick the line that matches your OS. Windows users: paste the install.cmd line, not the shell one.

# macOS or Linux
curl -fsSL https://clawmetry.com/install.sh | bash

# Windows (cmd.exe or PowerShell)
curl -fsSL https://clawmetry.com/install.cmd -o install.cmd && install.cmd

2. Run the daemon

clawmetry

The daemon auto-detects Deep Agents's local store and starts reading immediately.

3. Open the dashboard

Open http://localhost:8900. Switch the runtime chip to Deep Agents.

Start the free trial →

The one thing this catches that Deep Agents alone does not

Left alone, Deep Agents's failure mode is a nested sub-agent that spawns its own sub-agents and quietly outspends the parent by 10x. ClawMetry watches for it: cost per hour on the Deep Agents runtime, message-repeat detection, and per-session token-growth curves. Set a threshold, and the approval queue pauses before the next call. Kill switch (Pro) stops it.

All 19 runtimes ClawMetry observes

Deep Agents is one of 19. Free on every plan: OpenClaw, NVIDIA NemoClaw. Paid on Cloud Pro or Self-Hosted Pro: the rest.

OpenClaw NVIDIA NemoClaw NanoClaw PicoClaw Claude Code Codex Cursor Aider Goose opencode Qwen Code Hermes Pi Deep Agents (this page) n8n Antigravity GitHub Copilot Grok QM