Skip to content
Process Telemetry for Technical Hiring

Stop Interviewing Prompt Proxies.
Start Hiring Architects.

The Senior signal is dead. Promptster provides the telemetry you need to see the brain behind the AI.

HUMAN INPUT
ORCHESTRATION
ATTRIBUTED OUTPUT
# implement auth middleware
const token = req.headers
.authorization?.split(" ")[1]
if (!token) return 401
const payload = await
jwt.verify(token, SECRET)
// attach user to context
ctx.user = payload.sub
LLM Callanalyze intent
Read Toolsrc/middleware.ts
LLM Callgenerate implementation
Edit Toolauth/middleware.ts
Bash Toolpnpm tsc --noEmit
LLM Callreview + refine
export function authMiddleware(
req: Request,
res: Response,
next: NextFunction
) {
const auth = req.headers
.authorization
if (!auth) return res
.status(401).json({
error: "Unauthorized"
})
}
HUMAN
AI-GENERATED
MIXED

The Hidden Cost of Guessing

Every Wrong Hire Compounds

Technical interviews were designed for a pre-AI world. The cost of getting it wrong isn't just the hire — it's the months of lost velocity.

$0K

Average cost of a mis-hire at senior level

Salary, onboarding, lost productivity, severance

0.0 mo

Average time to spot a mis-hire

Lead time you can't get back

0%

Of developers use AI coding tools daily

The interview still tests for the pre-AI skill set

How It Works

Key → Candidate → Insight

Recruiter generates a key. Candidate runs one command. Promptster captures the session. Structured signal lands in your dashboard — no setup friction on either side.

03 · Insight

Insight Lands in Your Dashboard

Every decision event flows to your dashboard in real time. When the session ends, you get a structured report: percentile score, health indicators, a one-paragraph hiring brief, and auto-generated follow-up questions.

promptster report --session PST-3K9X-7FQM

Candidate Report · session PST-3K9X-7FQM

──────────────────────────────────────────────

Promptster Score78 / 100
PercentileTop 28%
Human Attribution73%

──────────────────────────────────────────────

Health Indicators

Loop detectedNone
First-turn resolutionHigh
AI over-delegationMild

──────────────────────────────────────────────

Hiring Brief

Strong decomposition + Redis decision well-reasoned.

Mild AI over-delegation on edge cases — probe further.

──────────────────────────────────────────────

Follow-up Questions

1. Why Redis vs. in-memory for this scale?

2. Walk me through the mutex decision.

01 · Recruiter

Recruiter Generates a Key

Create an assessment key (PST-XXXX-XXXX) in your dashboard. Send it via email, Slack, or your ATS. One key per assessment — no shared credentials.

promptster dashboard

Assessment Keys

RoleSenior Backend Engineer
Taskrate-limiter-v2

────────────────────────────────

KeyPST-3K9X-7FQM✓ copied

────────────────────────────────

Send to candidate via email, Slack, or ATS.

One key · one session · no shared credentials.

02 · Candidate

< 2 min setup

Candidate Runs One Command

The candidate installs promptster globally and starts their session with the key. A consent TUI shows exactly what will be captured before recording begins. Under 2 minutes from zero to live telemetry.

~/candidate-shell · live

npm install -g @promptster/cli

+ @promptster/cli installed globally

promptster start PST-3K9X-7FQM

Promptster — Consent

Capturing: prompts, tool calls, diffs,

and workspace file interactions only.

NOT capturing: keystrokes, clipboard,

screen recording, personal files.

[Y] Consent and begin session

Telemetry active · session PST-3K9X-7FQM

claude

Under 2 Minutes to Start

Three commands. Consent screen. Live telemetry.

1Install globally

npm install -g @promptster/cli

one-time · ~5 seconds

2Start with your key

promptster start PST-XXXX-XXXX

consent screen · then live

3Work normally

# open Claude Code or Cursor

telemetry runs in background

The signal you're missing

Three Ways to Evaluate. One That Actually Works.

You're not getting bad signal from your current process. You're getting no signal at all. Here's the difference.

Code output + pass/fail

Traditional Assessment

You see the final diff. You get a checkmark or a failing test. You have no idea whether a senior engineer wrote it in 20 minutes or a junior spent 3 hours copy-pasting from Stack Overflow with heavy AI.

  • No process visibility
  • Pass/fail only
  • AI indistinguishable

Video playback, keystroke heatmaps

Screen Recording

You get a video. You have to manually watch it. There's no semantic understanding — you can't query 'show me all decision points' or 'what was their exploration pattern.' It scales to exactly one reviewer-hour per candidate.

  • Manual review only
  • No semantic search
  • No attribution
Promptster

Annotated timeline · attribution · decisions

Promptster Telemetry

Every event is structured data. Attribution at the line level. Percentile rank across assessed candidates. Auto-generated hiring brief and follow-up probe questions. Consent-first — candidates see the exact capture scope before a single byte is recorded.

  • Line-level attribution + percentile rank
  • Health indicators + hiring brief auto-generated
  • Consent-first candidate experience

Feature Deep-Dive

Insight, Not Surveillance

Promptster captures the engineering process — the decisions that separate architects from auto-complete approvers.

promptster attribution --session PST-3K9X-7FQM

Attribution Report · session PST-3K9X-7FQM

───────────────────────────────────────

Total Lines Written247
Human-authored73%
AI-assisted27%

───────────────────────────────────────

AI suggestions accepted14 / 31
AI suggestions modified12 / 31
Prompt quality score8.4 / 10

───────────────────────────────────────

Breakdown

Human ██████████████████████████ 73%

AI-gen ██████████ 27%

Problem Library

Real Bugs. Real Signal.

We curate verified open-source issues from production TypeScript and JavaScript codebases. Candidates fix the actual broken commit — no toy problems, no LeetCode puzzles.

github.com/colinhacks/zod

abort:true skips inner validations on nested objects

warmup~30 min
TypeScriptparsingvalidation

github.com/fastify/fastify

routerOptions mutation causes shared state across instances

standard~60 min
Node.jsroutingencapsulation

github.com/trpc/trpc

Subscription backpressure drops messages under load

deep~90 min
TypeScriptWebSocketsstreaming

github.com/drizzle-team/drizzle-orm

Relation inference fails on composite foreign keys

standard~45 min
TypeScriptSQLORM

github.com/honojs/hono

Middleware chain short-circuits on async error boundary

deep~75 min
TypeScriptmiddlewareerror handling

Custom Import

Or import your own codebase

Point Promptster at any GitHub issue with a reproducible broken commit. We validate the SHA, generate a setup script, and add it to your private library. Assess candidates on the problems your team actually ships against.

Learn More

Trust & Privacy

Not Spyware. Here's the Diff.

Most HR tech earns the spyware label. Promptster captures process telemetry — the same data you'd put in a PR description — with explicit candidate consent and full transparency.

What We Capture

  • Prompts, tool calls & file interactions within the workspace
  • Shell commands run in your terminal within the workspace
  • File changes made in any editor within the workspace
  • Prompt strategy, iteration patterns, and revision history
  • Architectural decisions and tradeoff comments

What We Don't

  • Keystrokes, clipboard, or screen recordings
  • Commands or files outside the assessment workspace
  • Browser history or personal browsing
  • Webcam, microphone, or biometrics
  • Environment variables, shell history, or dotfile contents

Consent TUI

Before recording starts, candidates see a full-screen terminal UI listing every data type Promptster captures. They read it, confirm, and proceed. No fine print, no hidden telemetry.

Data Retention Controls

Configurable retention windows. Purge candidate data automatically or on request. GDPR-friendly by default.

Redaction Mode

Auto-redact secrets, tokens, PII, and API keys from session logs before they hit storage. Configurable regex rules. Coming soon.

On-Prem Deployment

Run the entire stack inside your VPC. Data never leaves your network. Available on Enterprise plans.

What you get

Four Signals. One Candidate. 90 Minutes.

Promptster generates structured telemetry that answers the questions your interview loop takes four rounds to ask.

Attribution

Human vs. AI vs. Removed

73% Human-authored
20% AI-assisted
7% Removed

session_4f8a · 247 lines

Efficiency

Prompt Efficiency Over Session

Score improves as candidate learns the codebase context

efficiency score (0–100)

Decisions

Decision Event Scatter

Each dot is a flagged decision moment, sized by impact severity

CriticalRejectionMinor

Orchestration

Score Breakdown · 6 Dimensions

Decomposition92
Tool Sequencing88
Integration85
Iteration Depth94
Prompt Quality84
Verification79

Sample data from a real assessment session · 47m 23s · 142 events captured

Pricing

Founding Pricing for Early Teams

Priced on reviewer seats and assessment volume — with lower early-team pricing while we earn trust, prove signal, and tighten the product.

MonthlyAnnualSave 20%

Starter

For teams running occasional technical hiring cycles.

$249 / monthly

  • 3 reviewer seats
  • 25 assessments / month
  • Full session replay
  • Attribution intelligence
  • Orchestration scoring
  • Email support
Get Started
Recommended

Growth

For growing teams with regular technical hiring volume.

Best for teams running 10–30 technical hires per quarter.

$699 / monthly

  • 10 reviewer seats
  • 100 assessments / month
  • Everything in Starter
  • Advanced analytics
  • Priority support
  • API access
Get Started

Founding members lock in this price.

Scale

For high-volume hiring with enterprise-grade requirements.

$1499 / monthly

  • 25 reviewer seats
  • 300 assessments / month
  • Everything in Growth
  • SSO / SAML
  • ATS integrations
  • Dedicated CSM
Get Started

Enterprise

For organizations with strict compliance, security, and on-prem requirements.

Custom

  • Everything in Scale
  • On-prem deployment
  • Custom compliance controls
  • Dedicated security review
  • Custom SLA
  • Volume discounts
Contact Sales

Ready to See How Engineers Actually Work?

Design partner slots are limited. Early partners get founding pricing, direct product input, and a dedicated onboarding session. We respond within 48 hours.

Design partner program — limited slots. Founding pricing locked in.