Ye tool kya hai?

Pheebs ko banaya hai Eversynced ne — aur ye ek open-source AI telemetry tool hai jo measure karta hai: tumhari team AI coding agents ke saath actually kaise kaam karti hai.

Aasaan bhasha me samjho: adoption dashboards batate hain “kisne AI tool use kiya” — lekin ye nahi batate ki session ke andar kya hua. Kaunsa model kis kaam par use hua? Kya bada model chhote kaam par chalaya gaya (budget waste)? AI ka likha code bina check kiye ship ho gaya ya pehle test chala? Pheebs inhi sawalon ke jawab deta hai.

Pheebs homepage — open source telemetry by Eversynced

Khaas features (jo ise alag banate hain)

1. Teeno bade coding agents me hooks

Pheebs Claude Code, Cursor aur Codex ke andar hooks lagakar baithta hai. Jab session start hota hai, prompt submit hota hai, tool call complete hota hai, model switch hota hai ya context compact hota hai — har ek cheez ek lightweight event ban jati hai. Kul 17 event types hain — session_started se lekar artifact_found tak.

Pheebs telemetry example — tumne kya kiya, aur Pheebs ne kya record kiya

2. Privacy-first: shape of the session, not its contents

Ye sabse serious design decision hai. Pheebs jo capture karta hai, wo ye hai:

  • Event type, durations, counts, models, trigger types
  • Prompt ka sirf character count (aur optional intent classifier label) — text kabhi nahi
  • Command strings (npm test jaise) process me padhkar tool_intent: test_run me convert — raw command kabhi store nahi

Aur jo capture nahi hota: source code, file contents, file paths, prompt text, naam, email. Identity sirf token ka ID hota hai — ya bina token ke git email ka truncated hash.

Pheebs privacy — kya capture hota hai, kya nahi

3. Local-first — tumhara data, tumhari machine

Har event pehle ~/.pheebs/logs me JSONL me save hota hai. Backend par kuch bhi bhejne ke liye tumhe khud base-url aur token dono set karne hote hain. Ek bhi unset hua, to sending ruk jati hai. Local JSONL hamesha durable copy rehti hai. Tum apna backend self-host kar sakte ho (repo me reference backend hai), ya Eversynced ka managed option le sakte ho.

4. AI Proficiency Model — data ko samajhne ka framework

Raw events ka stream confusing hota hai, isliye Pheebs ke saath AI Proficiency Model aata hai — data ko padhne ka structured tareeka:

  • Layer 1 (Repertoire): 6 competencies me 24 practices — Models, Artifacts (skills/sub-agents/slash commands), MCP, Evals (tests/lint/build), Context management, Orchestration
  • Layer 2 (Judgement signals): AI output verify hota hai ya nahi (verification coverage), engineers AI output ko challenge karte hain ya blind accept karte hain (pushback rate), follow-ups refine kar rahe hain ya breakage fix kar rahe hain (refinement-to-repair ratio), aur model fit (kaam ke size ke hisab se model)

5. Overspend ko dollars me price karta hai

Site par dikha example: ek team ke $32,400 ke monthly model spend me se $9,960 (31%) oversize models par ja raha tha. Pheebs model-fit gap ko directly dollars me price karta hai — ye woh line hai jo engineering manager ko budget meeting me kaam aati hai.

AI Proficiency Model — practice adoption aur judgement signals

Pricing — free me kitna milega?

  • Open source (self-hosted): 100% free — Apache-2.0 license, poora client, teeno agents. Backend tum khud run karo, data tumhare infrastructure me rahega. Koi account nahi, koi key nahi.
  • Managed (AI Enablement Assessment): 30-day telemetry sprint — Eversynced backend chalata hai aur proficiency reports/dashboards maintain karta hai, executive debrief aur improvement plan ke saath. Iski pricing site par publicly listed nahi hai.

Kaun use kare, kaun na kare?

  • Engineering managers & CTOs: Team me AI coding agents ka ROI samajhna ho, overspend rokna ho, ya quality gaps (bina test ke ship hota code) dhoondhna ho — ye tumhare liye bana hai.
  • DevOps/platform teams: Apna telemetry backend self-host karke JSONL data ko apne existing tools me query kar sakte ho.
  • Solo developers: Local mode free hai — apne khud ke AI usage patterns samajhne ke liye achha, lekin team-level insights nahi milenge.
  • Non-developers: Ye tool sirf code walo ke liye hai — marketers/content creators ke liye nahi.

Final verdict

Pheebs ek real problem ko solve karta hai jo har AI-first engineering team me hai: “Hum AI par itna kharch kar rahe hain, lekin andar kya ho raha hai pata nahi.” Privacy-first architecture genuine hai (open source = verify kar sakte ho), local-first design sensible hai, aur model-fit ko dollars me price karna smart business insight hai.

Aadha number isliye kata: sirf 3 agents supported hain abhi, aur naya product hai — ecosystem mature hone me waqt lagega.

Humari rating: 4.5/5. Agar tumhari team me AI coding agents seriously use hote hain — ye install karke dekhne layak hai, aur free hai.

Note: Ye first-look review official site (eversynced.com/pheebs) aur public GitHub repo (eversynced/pheebs) par based hai. Hum ise hands-on install karke testing ke baad update karenge.