session provision VPS · attach coding agent · record evidence
Agent-native workbench
See how engineers work with agents.
BrushPass gives teams disposable Linux workspaces with controlled AI access, terminal telemetry, file snapshots, and reviewable artifacts, so real engineering work can be evaluated, audited, and improved.
candidate@brushpass
ssh live
Candidate: Dalton Cole
Task: FizzBuzz API rescue
Active
Server IP
165.227.179.233
Commands
123456788
AI events
004444444
State
011122344
AI cost
$0.00$0.00$0.11$0.11$0.11$0.11$0.11$0.11$0.11
SSH connectedworkspace online
Task viewedTASK.md opened
AI-assisted build4 events · 18.4k tokens · $0.11
Tests passing8 tests passed
Runtime state changedapp listening on :8080
Endpoint verifiedFizzBuzz response HTTP 200
Git delta captured3 files · +91/-0
Documentation capturedREADME.md updated
ephemeral VPS
fresh Ubuntu per session
AI proxy
short-lived tokens
full timeline
shell, AI, files, ports
review artifacts
README, diffs, checks
Git-seeded tasks
repo, branch, or commit
How BrushPass works
Disposable servers, controlled agent access, reviewable evidence.
Start with a task. BrushPass provisions the workspace, proxies coding-agent traffic, records the work, and tears the box down when the session ends.
No provider keys on the VPS
Token scoped to one session
Git or empty workspace
Automatic teardown
Use cases
One workspace model, several review workflows.
Hiring assessments
Evaluate how candidates debug, deploy, use agents, and communicate technical tradeoffs.
Bug hunts
Seed a real repository at a branch or commit and watch how engineers investigate, patch, test, and document the fix.
Agent workflow audits
Understand how engineers use AI tools across real tasks without exposing provider keys.
Training and upskilling
Give teams realistic exercises and review the evidence behind their engineering decisions.
Vendor evaluations
Run short, contained work sessions for contractors or partners before granting broader access.
Field notes
Thinking out loud about agent-native engineering work.
May 25, 2026 · 6 min read
How BrushPass captures agent-native engineering work
A practical walkthrough of the disposable VPS, coding-agent proxy, terminal telemetry, git shadow, snapshots, and review evidence behind a BrushPass session.
May 22, 2026 · 4 min read
Evaluating agent-native engineers
Why technical interviews need to measure judgment, verification, Linux fluency, and agent workflow instead of only final code.
Security model
Built around disposable access and controlled AI traffic.
Ephemeral servers
Each session gets a fresh Ubuntu workspace with automatic teardown.
Short-lived AI tokens
Coding agents talk through BrushPass proxy tokens instead of raw provider keys.
Organization-owned keys
Provider credentials stay encrypted and scoped to the customer organization.
Budget and usage controls
Track token usage, estimated cost, and session-level AI activity.
Pricing
Simple platform access, metered workspace runtime.
Start with a flat organization subscription. Bring your own OpenAI key for AI usage, then pay only for the workspace runtime your sessions consume.
Team subscription
$250
/ month per organization
Includes the hosted review app, task templates, Git-seeded sessions, candidate portals, AI proxy controls, telemetry, review timelines, and teardown automation.
- Bring your own OpenAI key
- Anthropic Claude support coming soon
- Metered VPS runtime by workspace size
- Usage visibility for sessions, tokens, and infrastructure