Coding Agent Sentry Observability

One dashboard for your coding agents.

This tool watches local telemetry from Codex, Claude Code, and Pi, stores a private local history in SQLite, and can send clean usage traces to Sentry.

Use it to answer simple questions:

• Which agent is running right now?
• How many LLM calls happened?
• Which model is being used?
• How many tokens did we spend?
• What did it cost?
• Which tool calls or failures happened?

Screenshots

Sentry usage dashboard

Sentry traces view

What it does

Codex / Claude Code / Pi logs
        ↓
agent-vm bridge
        ↓
local SQLite memory database
        ↓
local dashboard + memory commands
        ↓
optional Sentry dashboards and traces

The local database is the source of truth. Sentry export is optional.

Supported agents

Agent	Source
Codex	~/.codex/logs_2.sqlite, ~/.codex/state_5.sqlite
Claude Code	~/.claude/projects/**/*.jsonl
Pi	~/.pi/suggester/logs/events.ndjson and project-local .pi/suggester logs

Install

git clone https://github.com/zonko-ai/coding-agent-sentry-observability.git
cd coding-agent-sentry-observability
./scripts/install.sh

The installer:

1. creates .venv
2. installs agent-vm
3. creates ~/.config/coding-agent-sentry-observability/env
4. imports the last 7 days of local agent history

To skip the first import:

AGENT_VM_INSTALL_SKIP_BACKFILL=1 ./scripts/install.sh

Start collecting live data

Run in the background on macOS:

agent-vm start-launchd

Or run in the foreground:

agent-vm bridge --loop

Check it:

agent-vm status

Local dashboard

agent-vm dashboard --port 8765

Open:

http://127.0.0.1:8765

This works without Sentry.

Sentry setup

Edit:

~/.config/coding-agent-sentry-observability/env

Set these values:

SENTRY_DSN=...
SENTRY_ORG=...
SENTRY_PROJECT=agent-vm-usage
SENTRY_PROJECT_ID=...
SENTRY_AUTH_TOKEN=...   # only needed to create/update dashboards

Apply the Sentry dashboard:

agent-vm sentry apply-dashboards

Backfill history into Sentry:

agent-vm backfill --minutes 10080

10080 minutes is 7 days.

Useful commands

Command	What it does
agent-vm status	Show config, service state, and recent usage
agent-vm start-launchd	Start the background collector on macOS
agent-vm stop-launchd	Stop the background collector
agent-vm bridge --loop	Run the collector in the foreground
agent-vm backfill --minutes 60	Import/export the last hour
agent-vm dashboard --port 8765	Start the local dashboard
agent-vm sentry apply-dashboards	Create or update Sentry dashboards
agent-vm memory search "query"	Search local agent memory
agent-vm memory context --cwd /repo --agent codex	Print useful local context for an agent

What gets tracked

• agent name: codex, claude-code, pi
• model name when available
• LLM calls
• token counts
• estimated or exact cost when available
• sessions and projects
• tool calls
• durations
• failures
• recent memory and session summaries

Privacy defaults

The default setup is local-first and text-minimizing.

By default:

• raw prompt text is not exported
• raw response text is not exported
• text previews are stored as length/hash summaries
• common secrets and emails are redacted
• local path tags sent to Sentry are hashed
• the local SQLite database stays on your machine

Read more in docs/PRIVACY.md.

Configuration

Main config file:

~/.config/coding-agent-sentry-observability/env

Common settings:

Variable	Default	Meaning
SENTRY_DSN	empty	Enables Sentry export
SENTRY_ORG	empty	Sentry org slug
SENTRY_PROJECT	agent-vm-usage	Sentry project slug
SENTRY_PROJECT_ID	empty	Numeric Sentry project id
AGENT_VM_MEMORY_DB	~/.coding-agent-sentry-observability/memory.db	Local SQLite database
AGENT_VM_POLL_SECONDS	15	Live collector poll interval
AGENT_VM_RECORD_MEMORY	1	Store normalized traces locally
AGENT_SENTRY_INCLUDE_TEXT	0	Opt into redacted text export

Full details: docs/CONFIGURATION.md.

Troubleshooting

Check whether the background collector is running:

agent-vm status

Watch recent bridge logs:

tail -f ~/Library/Logs/coding-agent-sentry-observability/bridge.out.log

Check errors:

tail -f ~/Library/Logs/coding-agent-sentry-observability/bridge.err.log

Send a test event:

agent-vm self-test

Run a dry-run backfill:

agent-vm backfill --minutes 5 --dry-run

Development

. .venv/bin/activate
python -m pytest
python -m agent_vm_observability status
python -m agent_vm_observability backfill --minutes 5 --dry-run

Docs

• Architecture
• Configuration
• Privacy

License

MIT. See LICENSE.