A platform for technical interviews and screening rounds. Interviewers and candidates meet in a room with video, a shared workspace running in a sandbox, and a whiteboard for design interviews. You can allow AI assistance, and you can see everything the candidate did.

Can candidates use AI?

Only if the interviewer allows it. Each interview is set to off, ask-only, or agent, and the interviewer can change it during the session. The candidate is told whenever it changes.

Does AI score or reject candidates?

No. Interviewers score with a rubric. The optional AI review is a summary with evidence, labeled advisory, and it gives no hire score. In screening, AI suggestions are labeled as suggestions and a person records every decision.

How do you deal with cheating?

Eyrie records sessions and notes unusual activity, such as large pastes or leaving the window, as information. An optional strict mode locks the workspace when a candidate leaves fullscreen or switches tab. No web tool can stop someone from using another device, so Eyrie shows you what happened and leaves the judgment to you.

Is candidate code sandboxed?

Each session runs in its own isolated environment with limited outbound network access. Your API keys never enter it and hidden tests are never inside it.

Interviews that look like the job.

Run live coding, system design, and screening rounds in a sandboxed environment. Allow AI when you want to see how candidates work with it. Your team makes every decision.

Start free Book a demo

No installs for candidates.

Senior Backend Engineer · Wallet service

Priya RamanLive32:14 / 60:00

AI: Ask-only18.4k / 100k tokens

WorkspaceResumeDatabase

Explorer

▾ src/

webhooks/handler.ts

ledger/transfer.ts

▾ tests/

double_spend.test.ts

ledger/transfer.ts

 1  export async function transferFunds(tx: Transaction, req: TransferRequest) {
 2    // Lock source account row to prevent double spend race
 3    const [account] = await tx.query(`
 4      SELECT id, balance, status
 5      FROM accounts
 6      WHERE id = $1
 7      FOR UPDATE
 8    `, [req.sourceAccountId]);
 9  
10    if (!account || account.balance < req.amount) {
11      throw new InsufficientFundsError(req.sourceAccountId);
12    }
13    await tx.query(`UPDATE accounts SET balance = balance - $1 WHERE id = $2`, [req.amount, req.sourceAccountId]);
14  }

Terminal · npm testExit code 1

2 passed· 1 failed: double_spend.test.ts (webhook idempotency race)

You

Mic on

Priya

Speaking

Live Controls

AI Mode

OffAsk-onlyAgent

Event Feed

14:02Priya asked: why is the balance check racy?

14:05Tests: 2 passed, 1 failed

14:09Hint sent: think about concurrent webhooks

Illustrative product screen with sample data.

Everything is recorded.

Replay the session file by file, with prompts, tests, and token usage.

AI only when you allow it.

Off, ask-only, or agent. Change it live. Candidates are always told.

People decide.

Scores come from your interviewers. AI only summarizes.

Puzzles don't show how people work.

Engineers work in real codebases, with real tools, and now with AI. An interview should show that, and give you evidence you can review and discuss as a team.

Aspect	A typical interview	In Eyrie
The task	An isolated puzzle	A real codebase with tests and a database
AI and tools	Banned, or used where you cannot see	Your rules for AI, with every prompt recorded
Evidence	Notes and memory	A replayable session with diffs and test results
The decision	Gut feel	Blind scorecards and a side-by-side debrief

The task

A typical interview:

An isolated puzzle

In Eyrie:

A real codebase with tests and a database

AI and tools

A typical interview:

Banned, or used where you cannot see

In Eyrie:

Your rules for AI, with every prompt recorded

Evidence

A typical interview:

Notes and memory

In Eyrie:

A replayable session with diffs and test results

The decision

A typical interview:

Gut feel

In Eyrie:

Blind scorecards and a side-by-side debrief

One room for the whole conversation

Video, a shared VS Code workspace, a terminal, and a resume viewer that only interviewers see. Add Postgres and Redis when the task needs a real database.

No installs for candidates
Pause, extend time, or send a hint without leaving the room
Interviewers watch the candidate's workspace read-only

Learn more

Room Tools · Interviewer Only

Priya Raman.pdfPage 1 of 2

Backend Engineer · 6 years

Fintech Platform · 2021 — Present

Led payments migration to transactional ledger architecture.

Cloud Infrastructure · 2018 — 2021

Maintained distributed webhook processing clusters.

Postgres 16Connected

Tables: accounts, transfers, webhook_events

SELECT id, balance FROM accounts WHERE id = 42;

idbalance

42$14,500.00

EXPLAIN: Index Scan using accounts_pkey on accounts (cost=0.28..8.29)

AI Permissions & Telemetry

Session AI PermissionConfigurable per interview live

OffAsk-onlyAgent

Token Budget42,180 of 100,000 tokens · $0.31

Prompt 7 · Agent turn2 files changed

"Write a test that reproduces the double spend"

Prompt 6 · Ask-only0 files changed

"Explain the isolation level guarantees of Postgres Read Committed"

Prompt 5 · Agent turn1 file changed

"Refactor webhook signature verification into standalone middleware"

Candidate view: "Your interviewer changed AI mode to Agent."Audited

Allow AI on your terms

Choose off, ask-only, or agent for each interview and change it while it runs. You see every prompt, every tool call, and the tokens and cost they used.

Candidates are told whenever the mode changes
Set a token budget for each session
The AI review is a summary with evidence, never a hire score

Learn more

Design interviews without a sandbox

A shared whiteboard and notes pane, with a rubric for requirements, data model, scaling, and trade-offs. Sessions start instantly because there is no environment to build.

Board snapshots appear in the replay
Freeze the board when time is up
Export the final diagram

Learn more

Organization Templates · Repository Ingestion

Wallet service: concurrent webhook handling

Ingested from github.com/internal/wallet-service.git

v3 · Published

Build & Validation Checklist

Dependencies installed (pnpm install completed in 4.2s)

Seed script ran (seeded 10 accounts, 50 transfer records)

Visible tests run (6 tests discovered)

Reference solution passes (6/6 tests passing)

Starting code fails 3 tests (candidate starts with failing tests)

Interview on your own codebase

Upload a project or start from a base environment. Eyrie builds it, checks that the tests behave as expected, and only then lets you publish. Hidden tests stay outside the candidate's environment.

Versioned, so edits never change a scheduled interview
Private to your organization
Preview it exactly as a candidate would

Learn more

See how they got there, not just the answer

Scrub through the session and watch the code, terminal, prompts, and test runs unfold. Totals for tokens, cost, and tests sit at the top of the results page.

Unusual activity is shown as information, never a verdict
Add notes at any timestamp
Optional strict mode locks the workspace if a candidate leaves it

Learn more

Panel Debrief & Scorecard Reconciliation

Rubric Dimension	Interviewer A	Interviewer B	Interviewer C
Debugging approach	4	3	4
Use of AICalibration spread	3	4	2
Communication	4	4	4

Panel ConsensusDecision: Advance to final round

Recorded by the hiring lead

Decide together, with evidence

Interviewers score on their own first, so nobody anchors on the loudest voice. The debrief shows every score per rubric item, highlights disagreements, and links straight to the replay.

Written descriptions for every score level
Calibration reports for admins
A person always records the final decision

Learn more

Screen at volume without losing the human.

Run a campaign for a role and invite candidates in bulk. You choose the kind of round that fits.

Async take-home.

Candidates work in a timed sandbox at any time in your window. Hidden tests grade the code automatically.

AI spoken screen.

An AI asks the same structured questions of every candidate. Candidates are told it is an AI and can ask for a person or type their answers.

Human-led screen.

A short live interview with one of your team, for roles where you prefer a conversation.

Screening Campaign · Senior Platform Engineer (Q3)

Candidate	Round	Status	Tests	Time	Decision
Priya Raman	Take-home	Completed	6/6	48 min	Advanced
Marcus Webb	Take-home	Completed	5/6	55 min	Hold
Aiko Tanaka	AI screen	Completed	n/a	14 min	Needs review
Daniel Okafor	Take-home	Completed	6/6	41 min	Advanced
Sofia Martins	Take-home	In progress	3/6	33 min	n/a
Lukas Vogel	AI screen	Invited	n/a	n/a	n/a
Hannah Kim	Take-home	Completed	4/6	59 min	Needs review
Omar Haddad	AI screen	Completed	n/a	16 min	Hold

Every advance or reject is recorded by a recruiter. AI suggestions are labeled as suggestions.

Learn more about screening

From invite to decision in four steps.

Create the interview.

Pick a template, set the AI mode and token budget.

Wallet serviceAsk-only

100k tokens · 60m

Invite the candidate.

Send one link. Candidates join in the browser.

Candidate join link

eyrie.io/join/tk_94f8

Interview together.

Talk on video, watch the workspace, and adjust the session live.

● Live 32:14

Review and decide.

Replay the session, score the rubric, and decide as a panel.

Concurrency handling4 / 4

Consensus: Advance

People decide.
AI assists.

Eyrie never rejects a candidate and never produces a hire score. When AI is used, candidates are told, and a person reviews the result.

How Eyrie uses AI

What AI does

Helps candidates, only when you allow it.
Summarizes a session with evidence for you to check.
Asks structured questions in an AI screen, with a human alternative.

What AI never does

Reject or advance a candidate.
Judge voice, accent, tone, or appearance.
Analyze faces or use webcam data.

Built for code you do not trust.

Read about security

A separate sandbox for every session.

Each candidate works in an isolated environment that is destroyed when the session ends. Outbound network access is limited.

Keys never enter the sandbox.

Your AI provider keys stay on the platform. The sandbox only holds a short-lived token for that one session.

Hidden tests stay outside.

Grading runs from the platform against a snapshot, so a candidate cannot read the answers.

Consent before recording.

Candidates see what is recorded and what signals are collected before they start, and their consent is saved.

Preparing for an interview? Practice in the same environment.

Try realistic rounds in the same workspace candidates use in real interviews. Free, with daily limits.

Start practicing What to expect in an interview

Pricing by sessions, not seats.

Starter

$0per month

Try Eyrie with a small monthly allowance.

Get started

Team

$149per month

Custom codebases, whiteboard, and panel scorecards.

Get started

Business

$449per month

Screening campaigns, strict mode, and calibration.

Get started

Enterprise

Custom

Volume, SSO, and a dedicated contact.

Contact sales

A session is one candidate in one room. Candidates never pay.

See full pricing

Common questions.

See all questions

Run your next interview in Eyrie.

Start on the free plan, or talk to us about your hiring process.

Start free Book a demo

Interviews that look like the job.

Everything is recorded.

AI only when you allow it.

People decide.

Puzzles don't show how people work.

One room for the whole conversation

Allow AI on your terms

Design interviews without a sandbox

Interview on your own codebase

See how they got there, not just the answer

Decide together, with evidence

Screen at volume without losing the human.

Async take-home.

AI spoken screen.

Human-led screen.

From invite to decision in four steps.

Create the interview.

Invite the candidate.

Interview together.

Review and decide.

People decide.AI assists.

What AI does

What AI never does

Built for code you do not trust.

A separate sandbox for every session.

Keys never enter the sandbox.

Hidden tests stay outside.

Consent before recording.

Preparing for an interview? Practice in the same environment.

Pricing by sessions, not seats.

Starter

Team

Business

Enterprise

Common questions.

What is Eyrie?

Can candidates use AI?

Does AI score or reject candidates?

How do you deal with cheating?

Is candidate code sandboxed?

Run your next interview in Eyrie.

People decide.
AI assists.