codesolara
// for engineering managers and recruiters hiring software engineers

engineering screens
you can defend.

Candidates work with AI, the way they will on the job. You get a scorecard where every claim cites its evidence — checked.

Codesolara opens a real editor, terminal and file tree in the candidate's browser, with an AI assistant — Solara — in the panel beside them. What they ask it, what they keep, and what they throw out becomes a scorecard that shows how they actually think.

request a walkthrough →see a sample scorecard
// no candidate account// runs in the browser// pay per attempt, seats free

// what you get back

A scorecard where every claim cites the moment it came from — and any claim whose evidence doesn't hold is withdrawn, not guessed.

A. Mehta · editorial-cms · 58 minborderline
AI fluencycounts for 62%7.5
highCaught an incorrect proposal before accepting it

Reverted the assistant's timezone change after noticing it dropped the DST offset, then asked for a failing test before re-attempting.

transcript · 14:22src/schedule.py:32–34probe · dst_boundary
midDirected the assistant with enough context to be useful

Early prompts named the symptom but not the surface; the assistant explored three files before reaching the scheduler. Later prompts were specific.

transcript · 03:10tool_use · 7 calls
Code qualitycounts for 38%6.0
lowLeft the feed path inconsistent with the fix

The scheduler was corrected; the feed still computes its own window and disagrees with it after a DST transition.

src/feed.py:88probe · feed_window · fail
not assessedHandled an empty feed without erroring

Withdrawn by the verifier: the cited probe exercised the scheduler, not the feed, so it cannot support this claim. Left out of the score rather than marked down.

verifier · citation does not hold
◆ every citation above was re-checked by a second passoverride →

Illustrative — not a real candidate.

how the scorecard is built →

// what the candidate sees

57s · product simulationrequest a walkthrough →

A real workspace. An AI pair. An engineer who reviews, corrects, and verifies the work. An evidence-backed scorecard.

Everyone your candidate competes against already uses AI. The old tests were built to catch it. We were built for a world where everyone has it.

// how it works

01

Send one link

Pick a task, invite a candidate. No account for them, nothing to install — the workspace opens in their browser.

02

They build, with Solara

A realistic task in a real editor and terminal, with an assistant that can read, write and run code beside them — and it asks before it runs a command.

03

You get the scorecard

The session becomes a structured read on the candidate, every claim citing the moment it came from. Automatically — no panel review.

the long version → what the candidate actually sees
// what you get
01 —

Real work, not whiteboards

A realistic task in a real editor, terminal and file tree — with Solara in the panel beside it.

02 —

You see the judgment calls

Solara changes the code; the candidate reads it, runs it, and fixes or undoes what is wrong. What they do next is the signal a diff alone can't give you.

03 —

The scorecard is the product

Every attempt becomes an evidence-backed read on the candidate, on the dimensions you chose to weigh.

04 —

Checked, then double-checked

A second pass verifies that the evidence behind each score actually holds. When it can't, the scorecard says so instead of guessing.

05 —

Explainable & defensible

Every score cites its evidence. Humans stay in the loop and can override anything — on the record.

06 —

Free for the whole team

Unlimited seats. Invite every reviewer. You pay when a candidate actually interviews.

what a scorecard says, and how it's built →
// the AI-free test
✗  Unaided coding under glass
✗  Treats AI use as cheating
✗  A puzzle, not the job
✗  Pass / fail, no story
// codesolara
✓  Real work with a real AI pair
✓  Measures fluency with AI
✓  A realistic task in a real environment
✓  A scorecard with the evidence behind it

The AI-free test is not the only alternative — here is how the other platforms handle AI, and what "fluency with AI" actually means if you are going to measure it.

// pricing

You pay per attempt.
Every seat is free.

No subscription, no per-recruiter tax, nothing that expires. Three depths of interview — they differ by how long the candidate gets and how much they can lean on the AI assistant before it runs out.

Screen

A first read

  • 60 minutes
  • AI assistant for targeted help
  • A focused, single-surface task
  • Full scorecard
Standard · where we'd start

The working interview

  • 60 minutes
  • AI assistant for a full working session
  • A realistic feature in an existing codebase
  • Full scorecard
Deep

The onsite replacement

  • 90 minutes
  • AI assistant to lean on throughout
  • An ambiguous brief with real trade-offs
  • Full scorecard
✓ unlimited seats✓ unlimited problems✓ credits never expire✓ no lock-in
talk pricing →

// we're setting per-attempt pricing with our first teams — ask and we'll quote you plainly

// questions
Isn't letting candidates use AI just cheating?+
The opposite. Pretending AI doesn't exist is the cheat. We measure the skill that matters now: getting great work out of AI — and catching it when it's wrong.
Is Solara a command-line agent?+
No. Solara lives in a panel next to the editor — the candidate talks to it in plain language while they work. It can read files, edit them and run commands in their sandbox. Its edits land in the files at once and show in Source Control; before it runs a command, it asks the candidate. They also have a real terminal of their own; the two are separate things.
How is this different from HackerRank or Codility?+
Most assessment platforms now let candidates use AI, and some grade it — HackerRank links each grade to the transcript excerpt behind it. What we add is the check: a second pass re-reads every citation and confirms it says what the score claimed. When it doesn't, that criterion is withdrawn rather than guessed, so a decision rests on evidence that has actually been verified.Every platform, compared →
So what are you actually measuring?+
How someone works: the quality of their judgment, how they direct and correct the assistant, and whether the result holds up when it's exercised. You choose the dimensions and how much each one counts. The scorecard, in detail →
Do candidates need an account?+
No. They get a link, do the interview in the browser, and you get the result. Nothing to install. What we tell candidates →
What counts as an attempt?+
One candidate starting one interview. If they drop off — closed tab, lost wifi — they rejoin the same session right where they left it, at no extra charge. You're billed once, per candidate.
Where does the candidate's code run?+
In a container created for that attempt and destroyed after it, as an unprivileged user, with no access to your data or anyone else's. Security, in detail →
When can we start?+
Now. Leave your work email below and we'll reply to set you up. On your side, setup is creating a role and sending invites — nothing to install.
The way you hire engineers
just changed.
Send one link. Get back a scorecard you can defend.
javascript is required to request a walkthrough.