Local validation • Local by default • Project-scoped access

Pay only for scale, not for POCs.

Validate the idea first. Scale what works.

Designed for creators, founders, and small engineering teams: choose a project directory, connect your Codex account, run a small real test, inspect whether it worked and what it used, and decide whether the idea deserves production investment.

Codex usage itself follows your personal Codex account quota; Sidecar eliminates the upfront cost, complexity, and risk of building surrounding infrastructure before you even know if the workflow is viable.

rocket_launch Try your first POC science See how validation works

Runs natively on macOS & Windows • No local server configuration required

< 2 min

to first HTTP 200

0 infra

cloud footprint to start

4 steps

from install to test

100% local

gateway & dashboard

Sidecar Desktop
Daemon active

Project-Scoped Key

local-development ● ACTIVE

60 req/min limit • Optional expiry

Bound Workspace

folder ~/Projects/client-portal

Folder sandbox verified

Codex Auth Session

verified_user Connected

Direct upstream session

send Live Handshake Test
READY
Payload codex-mini (example) 
{
  "model": "gpt-6-astra",
  "messages": [
    {
      "role": "user",
      "content": "Say hello"
    }
  ]
}
Connection successful
200 OK
Latency (example): ~412ms
Tokens used (example): 32 total (14 / 18)
Auth Type: Bearer (local-development)
Daemon port: 3081 / loopback ok
check_circle Codex accepted request & returned valid completion
shield The desktop gateway and dashboard run locally by default. Request history keeps metadata only. Inspect full workflow arrow_forward
Fast & Frictionless Verification

See if it works before you scale it.

Setting up a new AI workflow traditionally means provisioning complex cloud gateways, wiring webhooks, managing hosted secrets, and hoping nothing breaks. Sidecar gets you to a confirmed HTTP 200 in under two minutes with zero infra footprint.

call_split Click a step to sync the dashboard mockup above.
Know Before You Commit

Understand cost, latency, and viability in real time.

When proving an AI feature, the biggest unknowns are token usage, unexpected response latency, and cryptic local connection errors. Sidecar surfaces the exact telemetry you need to evaluate if an idea is ready for production.

list_alt Recent Requests
Live Rate limit: 60 req/min
200 OK

IDE inline query

Codex model (example) • 1 min ago

412ms (example)

28 tokens (example)

200 OK

Prototype test completion

Selected model • 4 mins ago

1,240ms (example)

542 tokens (example)

429

Rapid script loop throttled

Safety bucket preserved quota

12ms (example)

0 tokens billed

200 OK

Local connection verification

handshake-ping • 14 mins ago

210ms (example)

32 tokens (example)

Volatile metadata only • Prompts never stored In-memory session
vital_signs System Diagnostics
All checks passed
check_circle Port Binding (3081)
LISTENING
check_circle Workspace Access
BOUND & VERIFIED
check_circle Codex Authentication
AUTHENTICATED
check_circle Local Safe Key State
ACTIVE & ENFORCED

Plain English Diagnosis

If a local port conflict occurs, your key expires, or upstream auth needs renewal, Sidecar identifies the root cause and provides a clear resolution path.

Zero Lock-in Integration

Use it where you already work.

Sidecar acts as an OpenAI-compatible local API endpoint. Connect your existing editors, scripts, or prototype frontends to validate ideas rapidly.

code

VS Code

Configure your standard AI extension to point directly to Sidecar's local endpoint for interactive chat and completion testing.

http://127.0.0.1:3081/v1/chat/completions
terminal

Cursor & Cline

Use standard OpenAI-compatible API base URL and your scoped safe key for seamless inline editor queries and suggestions.

Bearer key authentication
web

Browser Prototypes

Test client-side JavaScript experiments against local AI streaming without exposing upstream credentials in browser devtools.

Origin-checked local access
bolt

cURL & Scripts

Run deterministic test scripts or benchmarks locally before building any complex deployment pipelines.

curl http://127.0.0.1:3081/v1/models
info
Current Scope Note: Sidecar is optimized for standard OpenAI-compatible text completions and chat streaming. Client tool calls, VS Code Agent mode, and standard image uploads are not supported yet.
Local & Sovereign

Private by default.

When evaluating early ideas, you should never have to compromise confidentiality or send unvetted workspace data through unfamiliar third-party proxies.

computer

Runs on your computer

The desktop gateway and dashboard run locally on your computer by default. Traffic dispatches upstream straight to Codex.

folder_limited

Strict folder boundaries

Authorized tools are confined strictly to your chosen project workspace. Your personal files and system paths remain unreachable.

key_vertical

Project-scoped keys

Project-scoped API keys with rate limits and optional expiry. Keys are securely stored and verified on your local gateway.

memory

In-memory metadata only

Request history stores volatile metadata only (timestamp, model, status, latency, token count) in local memory and never stores prompts, completions, or credentials.

PLANNED, NOT AVAILABLE YET

On our roadmap

Upcoming features focused on making local AI validation even more frictionless.

Broader provider adapters

Connecting additional AI accounts with the same local boundary safeguards.

Standalone background daemon

Quiet background operation that activates whenever your IDE queries the local port.

One-line automated installers

Streamlined single-command installer packages for immediate desktop onboarding.

Deeper IDE extensions

Direct one-click connection extensions for auto-configuring editor base URLs.

FAQ

Quick answers before you start.

The questions that come up most before the first install.

shield_lock Where does my data and prompt history go?
Everything runs locally by default. Requests route through the Sidecar gateway on your machine, request history keeps metadata only, and prompts are forwarded to Codex as a direct account session — Sidecar does not host, log, or store your conversation content.
payments Is there a usage quota or do I pay per request?
Sidecar itself is free to use; it adds no per-request fees. Your personal Codex account supplies the model entitlement, so billing follows your own account's plan. Rate limits protect your quota with a configurable safety bucket.
bolt Which models does Sidecar work with?
Sidecar dispatches to your linked Codex session, so the models available from your Codex account are what you can test — no extra setup per model. Model telemetry (latency, token counts) is reported per request in the dashboard.
support_agent What if I need help or hit a connection error?
Open an issue on the Sidecar GitHub repository with your OS version and the test output. For quick checks, run the dashboard's "Test Connection" — it shows whether the local port, upstream handshake, or auth session is the blocker.
Local companion for fast AI validation

Ready to validate your AI idea without infra overhead?

Download Sidecar for macOS & Windows, connect your Codex account, and test your first proof of concept locally today.

rocket_launch Try your first POC View live test

Sidecar for macOS & Windows

Local AI gateway • Free

download Download