PromptOS
Workspace controls

Settings

Configure provider access, subscription intent, memory, privacy controls, safety guardrails, and execution modes.

API Keys

The backend reads provider keys from the environment. No keys = free offline mode.

Copy .env.example to .env at the backend root. Restart the server after changing.

Free model connector

Choose how the Free plan reaches no-cost models before any Hybrid or Premium paid source is used.

Local PC

This selects the model reference the backend should use for FREE_LOCAL_MODEL.

Llama 3.1Recommended defaultSaved

Balanced writing, analysis, and everyday prompt work.

Install command
ollama pull llama3.1
Backend env
FREE_LOCAL_MODEL=ollama:llama3.1

How to change the local model

  1. 1Install the model on this PC: ollama pull llama3.1
  2. 2Set FREE_LOCAL_MODEL=ollama:llama3.1 in backend\.env.
  3. 3Keep OLLAMA_BASE_URL=http://localhost:11434 unless Ollama runs somewhere else.
  4. 4Restart the backend so the router reads the new model.
  5. 5Run a prompt in Studio. If the local model is offline, the router falls back instead of failing.
Fallback order
1 LocalOffline engine

Free runs try the local open-source model first, then configured free connectors, then the deterministic offline engine.

Privacy & limits

Local answered prompts stay on this PC. Cloud fallback may send the prompt to the configured free connector, so keep the toggle off for private-only work.

Backend note: the router already reads FREE_LOCAL_MODEL and OLLAMA_BASE_URL. Persist this UI to backend settings when runtime configuration endpoints are added.

Budget & Subscription

Select your preferred usage plan. Upgrade to access premium models globally.

Context memory

Facts and style preferences injected into every prompt. The engine also learns these from your feedback.

Sign in to manage context memory — durable facts and style preferences the engine injects into every prompt.

Security & privacy

Encryption, your audit trail, and data controls.

Your data is encrypted & private

Passwords are hashed with PBKDF2-HMAC-SHA256 (200k iterations, per-user salt) — never stored in plain text. Traffic is served over HTTPS/TLS, and data is encrypted at rest. Secrets in your prompts are masked before any external model call.

Sign in to view your security audit trail and manage your data.

Safety guardrails

Automatically injected based on detected domain. Cannot be disabled.

medical

Frame as general information; recommend consulting a professional.

financial

Illustrative planning aid, not financial advice.

minor

Age-appropriate content only; no unsafe material.

Execution modes

ModeCallsUse
multi_agent7 chainedHighest quality, full trace
hybrid1 + 3 stagesFaster, cheaper, less granular
fast1 structuredLowest latency, simple queries
offline0 — rule-basedInstant, free, unlimited