--- title: "Monitor coding agents | Grafana Cloud documentation" description: "See usage, cost, tokens, and tools across Claude Code, Cursor, Codex, and every other coding agent your team runs." --- > For a curated documentation index, see [llms.txt](/llms.txt). For the complete documentation index, see [llms-full.txt](/llms-full.txt). # Monitor coding agents Every person on your team runs several coding agents, and each one keeps its own usage in its own silo. Nobody sees what the agents cost, how many tokens they burn, or which tools they call. Agent Observability puts all of it in one view, with no application code changes. ## Before you begin - Install the `agento11y` CLI on your machine. On macOS, use Homebrew (`brew install grafana/grafana/agento11y`). On Linux and macOS, you can also use the install script. For instructions, refer to [Set up and monitor coding agents](/docs/grafana-cloud/observe-and-act/agent-observability/get-started/instrument-coding-agents/#before-you-begin). - To share sessions with Grafana Cloud, ensure Agent Observability is enabled on your Grafana Cloud stack and your role has data-read permission. ## See all your agents in one view Install the integration once and every supported agent reports into the same place: Claude Code, Cursor, Codex, Copilot CLI, OpenCode, Pi, and more. Refer to [Supported coding agents](/docs/grafana-cloud/observe-and-act/agent-observability/get-started/instrument-coding-agents/#supported-coding-agents) for the full list. ## Start locally, share when you’re ready You see everything on your own machine first, and you choose what leaves it. [](/media/docs/agent-observability/agento11y-how-it-works.png) ### Review sessions in the local app The local app runs on your machine at `http://localhost:8765`. It doesn’t need an account or network access. It keeps full conversations, tool calls, tokens, and estimated spend as plain files on your disk, so you can see what your agents do before you share anything. Open `http://localhost:8765` in your browser, or run `agento11y local open`. Manage the app with `agento11y local start`, `status`, `stop`, or `restart`. The app has four pages: **Sessions**, **Analytics**, **Security**, and **Settings**. ### Inspect sessions and spend Use **Sessions** to find any conversation. Search prompts, responses, and tools, and filter by workspace, agent, model, or status. Each row shows the first prompt, workspace, agent, cost, tokens, duration, call count, and models, so you can spot an expensive or long-running session at a glance. Open a session to read the full thread. [](/media/docs/agent-observability/agento11y-local-app-sessions.png) Summary cards above the list show sessions, cost, total tokens, input cache hit rate, model calls, and errored sessions for the selected range. The app starts with sessions captured after you install `agento11y`. To add earlier sessions, use the banner on **Sessions** or go to **Settings** > **History**. Use **Analytics** to measure usage in cost or tokens over the last 30 days. Review estimated spend, total tokens, input cache hit rate, and model calls with the change from the previous period. Break usage down by workspace and model, and compare blended cost per million tokens across models. [](/media/docs/agent-observability/agento11y-local-app-analytics.png) Estimated spend is a partial estimate. The app excludes usage from models that it can’t price. Scroll down on **Analytics** to see which sessions led to merged work. Break usage down by branch and merge status, and see which languages your tokens went to. [](/media/docs/agent-observability/agento11y-local-app-analytics-merge.png) ### Stop risky actions with local guards Use **Security** to turn on guard packs that block tool calls or redact their arguments before they run. Guards run on your device and work offline. Rules are saved in `~/.config/agento11y/guards.toml`. The following table lists each guard pack, when to choose it, what it covers, and what it doesn’t cover: Expand table | Pack | When to choose | What it covers | What it doesn’t cover | |--------------------------|------------------------------------------------------------------------|------------------------------------------------------------------------------------|------------------------------------------------------------------------------------| | **Secret redaction** | Choose when tool calls might expose credentials. | Redacts API keys, tokens, and private keys in captured tool calls. | Unstructured sensitive text or custom secret patterns not defined in the rule set. | | **File safety** | Choose when you want to protect sensitive configuration files. | Blocks reading and writing `.env` files across file and shell tools. | Files outside `.env` patterns unless configured in custom rules. | | **Git safety** | Choose when you want to protect against lost work or history rewrites. | Blocks git commands that discard uncommitted work or rewrite commit history. | Normal git operations such as staging, committing, or branch creation. | | **Destructive commands** | Choose when you want to prevent accidental filesystem deletion. | Blocks recursive deletes such as `rm -rf`. | Non-recursive deletions or single-file removal commands. | | **Permissions** | Choose when you want to prevent unauthorized permission escalations. | Blocks recursive `chmod` and `chown`, and `chmod 777` on `/` or `$HOME`. | Targeted permission changes on non-root or non-home paths. | | **Disk wipe** | Choose when you want to protect storage drives and partitions. | Blocks irreversible disk operations such as `dd` to devices, `mkfs`, and `wipefs`. | Standard file-level modifications or non-raw block device writes. | [](/media/docs/agent-observability/agento11y-local-app-guards.png) Local guards are defense in depth, not a sandbox. Test your rules offline with `agento11y guards test`. ### Choose what you share When you connect to Grafana Cloud, the content capture mode controls what leaves your machine: Expand table | Mode | When to choose | What is shared | What stays local | |---------------------------|--------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------|----------------------------------------------------| | `metadata_only` (default) | Choose when privacy policies require conversation text to remain on your machine. | Model, tokens, tool names, timing, and cost | Prompts, responses, and tool arguments and results | | `no_tool_content` | Choose when you want to review conversation messages while keeping tool arguments and outputs private. | Conversation messages and metadata | Tool arguments and results | | `full` | Choose when you need full debugging of prompts, responses, and tool execution. | Everything, including messages, thinking, system prompts, and tool input and output (secrets are automatically redacted) | Nothing | ## Get a team view in Grafana Cloud In Grafana Cloud, navigate to **Agent Observability** and select **Coding agents** to view sessions, estimated cost, tokens, active repositories, and active users across your team. [](/media/docs/agent-observability/agento11y-cloud-app-breakdown.png) Use the breakdown picker on the usage chart to group sessions, tokens, or cost by dimension. To see who and what drives spend, enable automatic tags for `user`, `repo`, and `branch`. Then sort repositories and branches by estimated cost. [](/media/docs/agent-observability/agento11y-cloud-app-more-breakdown.png) Because agent data lands as standard metrics, you can also build alerts and reports in Grafana for spend or token spikes. For attribution details, refer to [Attribute cost to a user, repository, or branch](/docs/grafana-cloud/observe-and-act/agent-observability/get-started/instrument-coding-agents/#attribute-cost-to-a-user-repository-or-branch). ## Get started in three commands Run the following commands to install `agento11y`, authenticate with Grafana Cloud, and launch an agent: Bash ![Copy code to clipboard](/media/images/icons/icon-copy-small-2.svg) Copy ```bash brew install grafana/grafana/agento11y agento11y login agento11y claude ``` Replace `claude` with `codex`, `pi`, or another supported agent. For other platforms, credentials, and unattended rollouts, refer to [Set up and monitor coding agents](/docs/grafana-cloud/observe-and-act/agent-observability/get-started/instrument-coding-agents). ## Next steps - [Set up and monitor coding agents](/docs/grafana-cloud/observe-and-act/agent-observability/get-started/instrument-coding-agents) - [Monitor agents with analytics](/docs/grafana-cloud/observe-and-act/agent-observability/observe-and-troubleshoot/dashboards) - [Optimize cost and performance](/docs/grafana-cloud/observe-and-act/agent-observability/observe-and-troubleshoot/cost-optimization)