Agent workflow telemetry
Hooks into Claude Code, Cursor, and Codex to record session starts and ends, prompts, model changes, tool calls, failures, compaction, background tasks, and related workflow events.
Pheebs
Pheebs is an open-source telemetry tool for understanding how developers use Claude Code, Cursor, and Codex. It records lightweight workflow signals to help teams analyze model spend, verification habits, rework, and adoption without storing code or raw prompt text in normal event records.
Pheebs overview
Pheebs is an open-source telemetry tool from Eversynced for measuring how developers work with AI coding agents. It connects through hooks to Claude Code, Cursor, and Codex, recording interaction signals about sessions, models, prompts, tools, verification, and orchestration rather than storing the contents of the work.
The tool is intended to help teams understand model spend, workflow proficiency, and adoption of agent capabilities. Pheebs can operate locally or send events to a self-hosted or Eversynced-managed backend, where complete session data can be used to examine model fit, verification habits, rework, and enablement patterns.
Key features
Content-minimizing capture
Stores lightweight fields such as event type, durations, counts, model selection, trigger types, prompt length, and optional intent labels instead of the underlying interaction text.
Proficiency and spend analysis
Uses task sizing, model choice, effort settings, verification activity, repair prompts, and related signals to evaluate whether model capacity matched the work and estimate potential spend gaps.
Local-first event pipeline
Keeps a durable local JSONL copy and can forward events to a configured backend without blocking the coding agent; transport failures are handled asynchronously and are not retried by the client.
Backend flexibility
Supports a self-hosted or Eversynced-managed backend through a base URL and developer token, with documented routes for ingestion, validation, optional prompt classification, insights, and OTLP passthrough.
OpenTelemetry support
Configures native OpenTelemetry export for Claude Code and Codex through the Pheebs proxy, while hooks provide the main event capture path for all supported agents.
Use cases
Find avoidable model spend
Review model selection against task scope, effort settings, and repair prompts to identify sessions where a larger model was used than the work required or where an under-powered choice created rework.
Improve verification practice
Measure whether AI-generated changes were followed by tests, typechecks, builds, or review passes, helping teams see where verification is present or missing in the agent loop.
Evaluate agent enablement
Track recurring use of skills, slash commands, sub-agents, context files, MCP connections, and other agent configuration to determine which enablement investments are actually adopted.
Study team workflows
Use session, orchestration, compaction, background-task, and tool-failure signals to understand how developers structure AI-assisted work across a team without collecting project contents.
Pros and Cons
Pros
- Open-source client with local-only operation available by leaving the backend endpoint unset.
- Captures workflow metadata without storing source code, file contents, file paths, or raw prompt text in the normal event record.
- Supports both self-hosted and Eversynced-managed backend arrangements.
- Provides a documented backend contract and OpenTelemetry path for Claude Code and Codex.
- Connects telemetry to model-fit, verification, rework, and adoption analysis rather than reporting model usage alone.
Cons
- The public material does not specify pricing or provide a complete description of managed-service plans.
- Optional prompt classification sends raw prompt text to the configured backend; it requires explicit configuration and can be disabled.
- The tool is designed around three named coding agents, and the available evidence does not establish support for additional agents.
FAQ
How is Pheebs installed?
Install Pheebs globally with npm install -g pheebs, then run pheebs init. The interactive setup detects Claude Code, Cursor, and Codex, configures selected agents, and can be verified with pheebs doctor.
Does telemetry leave the developer's machine?
By default, Pheebs writes events to a local JSONL log at ~/.pheebs/logs/. Events are sent to a backend only after both a base URL and token are configured. A blank endpoint keeps the installation local-only.
What information does Pheebs capture?
Pheebs records interaction signals such as event types, durations, counts, models, trigger types, tool intents, and prompt length. It does not store source code, file contents, file paths, directory structures, prompt text, or command strings as raw text. Optional prompt classification sends prompt text to the configured backend and stores only the returned label.
Can teams use their own backend?
Pheebs can use a self-hosted backend or a backend managed by Eversynced. The client is backend-agnostic and uses a base URL and per-developer token; the documented contract includes ingestion, token validation, optional prompt classification, and an OTLP passthrough.
Which coding agents does Pheebs support?
Pheebs is designed for Claude Code, Cursor, and Codex. Claude Code and Codex can also export native OpenTelemetry metrics and logs through the Pheebs proxy; Cursor has no native OpenTelemetry export configured by Pheebs.
Quick Facts
- Category
- Developer Tool
- Product type
- Open-source AI telemetry
- Supported agents
- Claude Code, Cursor, Codex
- Install
- npm install -g pheebs, then pheebs init
- Storage
- Local JSONL by default; optional backend delivery
- Backend options
- Self-hosted or Eversynced-managed
Pheebs Alternativen
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
Epherem
Epherem is a static security scanner for web apps built with AI coding tools such as Lovable, Bolt, v0, Replit, Cursor, and Claude Code. It checks source files and dependencies for common security mistakes, then provides categorized findings and suggested fixes without running the app.
PromptScout
PromptScout tracks how ChatGPT, Gemini, Google AI Overviews, and Perplexity mention your brand or competitors, then pairs those results with source analysis and website audits. It helps teams decide what to fix in content, positioning, or site readiness next.
SaveMRR
SaveMRR is a Stripe retention tool for SaaS teams that scans billing data for churn and MRR leaks, then automates recovery through dunning, cancel-save offers, win-back emails, and onboarding nudges. It is built for founders and bootstrapped teams using Stripe.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
Hype
Hype is a web tool for finding trending YouTube topics by category, time range, and scoring mode. It helps creators spot emerging ideas, inspect source videos, and decide what to cover next.