Pheebs: An Open-Source AI Telemetry Tool That Quantifies How Engineers Collaborate with AI Coding Assistants

When AI Coding Becomes Routine, How Do You Measure Its Real Value?

As AI coding assistants like Claude Code, Cursor, and Codex rapidly embed themselves into developers' daily workflows, a widely overlooked question has come to the surface: how are teams actually using these tools? Usage frequency, interaction patterns, depth of collaboration — for most teams, this data is essentially a black box. Pheebs, an open-source tool from the Eversynced team, is aimed squarely at this blind spot, offering a quantifiable standard for measuring "how engineers and AI truly collaborate."

After launching on Product Hunt, the product received 75 upvotes and ranked 8th for the day, tagged under open source, artificial intelligence, GitHub, and data analytics, with Davi Prates listed as the Maker.

What Pheebs Does

Pheebs positions itself as an open-source AI telemetry tool. Its approach is notably elegant: it hooks silently into popular AI coding environments — Claude Code, Cursor, and Codex — capturing lightweight interaction signals quietly in the background.

The team uses a precise phrase to define its data boundaries: it captures "the shape of the session, not its contents." In other words, Pheebs is interested in the structural characteristics of your AI interactions — how long a session lasts, how many turns it involves, which tools are used, how usage patterns are distributed across team members — not the specific content of what you and the AI discussed or what code was written.

This design is especially critical in an era of growing data privacy concerns. For enterprises and teams that want visibility into AI tool adoption and ROI without exposing sensitive code or business logic to a third-party tool, Pheebs' "shape, not content" approach hits exactly the right balance.

Telemetry is an automated remote data collection technology originally used in aerospace and industrial control systems, and now widely applied in software engineering to monitor application performance, user behavior, and system health. In the developer tooling space, IDEs like VS Code and JetBrains have built-in telemetry modules that collect anonymous usage statistics such as feature frequency and crash reports. Pheebs extends this concept to the usage layer of AI coding assistants. Hooks are extension points that automatically execute preset logic when specific events are triggered. Claude Code, Cursor, and similar tools expose these interfaces, allowing third-party tools to inject custom logic when a session starts, ends, or a specific action occurs — enabling Pheebs to collect data non-invasively without modifying the host tool's code.

Why Teams Need AI Usage Metrics

Over the past year or two, nearly every engineering team has been paying for AI coding assistants, yet few can answer the question managers care most about: what has this investment actually delivered?

Traditional developer productivity metrics — code commits, PR counts — are losing relevance in the AI era. When a significant portion of code is generated or assisted by AI, raw output numbers no longer reflect real engineering behavior. Teams need new dimensions of observation to answer questions like:

• Which engineers have genuinely integrated AI into their workflow, and which are only occasionally dabbling?
• What are the adoption rates of different AI tools (Claude Code vs. Cursor vs. Codex) within the team?
• Are AI collaboration patterns healthy, or are there signs of over-reliance or inefficient loops?

Pheebs provides exactly this kind of behavioral-layer data foundation. It doesn't hand you efficiency conclusions directly — instead, it gives teams the raw interaction telemetry signals to interpret AI tool adoption on their own terms.

In traditional developer productivity frameworks, DORA metrics (deployment frequency, lead time for changes, change failure rate, time to restore) and the SPACE framework (Satisfaction, Performance, Activity, Collaboration, Efficiency) have been the dominant measurement systems. However, both frameworks were built on the assumption that code is produced independently by humans. As GitHub data shows that more than 40% of new code is now written with the assistance of tools like Copilot, metrics like lines of code and commit frequency have become far less correlated with individual engineering capability — and can even be misleading, since AI-generated boilerplate inflates surface-level output numbers. This context explains why "AI collaboration behavior itself" needs to be observed independently: what team leaders truly need to understand is no longer just how much was produced, but whether the quality and dependency patterns of engineer–AI interactions are evolving in a healthy direction.

Why the Open-Source Route Matters

Notably, Pheebs has chosen the open-source model. For a data-collection tool, open source means the collection logic is fully transparent — teams can audit the code to confirm exactly what is and isn't being captured. This is an important trust prerequisite for enterprises deploying telemetry tooling.

Open source also lowers the barrier to adoption. Teams can customize the dimensions they track to fit their own needs, or pipe the data into existing analytics pipelines, without being locked into a closed SaaS platform. For engineering teams that value data sovereignty, this is a compelling proposition.

A Measured Take: Still Early Days

Based on publicly available information, Pheebs is still at a very early stage. Product Hunt shows only 2 comments, and the product's specific metric framework, visualization capabilities, and depth of compatibility with different AI tools all remain to be validated.

"Quantifying AI collaboration" is an extremely valuable direction, but it also faces real challenges: how to define meaningful metrics, how to prevent the data from being misused as a surveillance tool for individual engineers, and how to derive genuinely insightful conclusions from the abstract layer of "session shape." These are questions Pheebs will need to answer going forward.

For teams that are scaling their use of AI coding assistants and want to build data-driven decision-making, Pheebs is worth adding to the watchlist as an early-stage open-source option.