PromptParle frequently asked questions

PromptParle is an AI context optimization gateway. It sits between your desktop tools and providers like OpenAI, Claude, Gemini, and Grok. It thins noisy context, keeps the useful signal, helps mask secrets, and routes cleaner prompts so you pay for less filler.

Large logs, docs, and code dumps burn tokens and bury the ask. Users hit plan limits mid-work. AI vendors bill on volume, they are not built to shrink your spend. PromptParle sends leaner, higher-signal context while you keep the models and accounts you already pay for.

You choose OpenAI, Claude, Gemini, or Grok models your key allows. PromptParle reduces bloated context so flagship models cost less per useful turn. Completions stay live from your provider every request.

Agents and multi-step local workflows can spread work, but every hop can still ship a fat context window. PromptParle optimizes the shared problem underneath: tokens that should never have hit the meter.

Often yes. Many free and mid-tier plans cut you off with “you’ve reached your max.” Stripping bloated tokens means fewer tokens per request, which can delay that wall or stop it for some workloads. Results depend on how noisy your context was.

Tokens are their revenue model. Larger context windows and heavy usage grow that business. PromptParle exists for the buyer side: keep the model quality you need, pay for less noise.

Neither. It is not its own foundation model. Chat runs in the free desktop client on your machine. Models are provided by the AI vendors you configure with your own API keys.

The portal handles invitations, plan, and desktop license keys (pp_live_…), plus optional usage stats. Day-to-day chat, optimize, provider keys, workspace, Git, and SSH run in the local desktop client (pp) on your PC.

A 1-5 control that trades fidelity for token savings: 1 Max Fidelity, 2 High Fidelity, 3 Optimized, 4 High Savings, 5 Max Savings. Lower numbers keep more of the original text; higher numbers aim for more reduction on noisy material. Typical bands: ~0-10% at dial 1 up to ~50-80% at dial 5 on noisy packs. Unique prose may show little or no reduction.

Job-oriented presets such as general, developer, security-review, log-analysis, documentation, and executive-summary. They bias what kinds of content to prefer keeping. They are not a guarantee of a fixed savings percentage.

OpenAI, Anthropic (Claude), Google Gemini, and xAI Grok when you supply a valid provider API key. Availability of specific models depends on your provider account and key permissions.

PromptParle is not a foundation-model trainer. Whether a third-party provider trains on API traffic is governed by that provider’s terms and your agreement with them. Use their enterprise/API settings if you need stricter data handling.

Registration is open and free. Create an account at https://promptparle.com/register with email (a quick verification link) or Google/GitHub. Then create a desktop license key (pp_live_) for each machine you install on.

No. Signup is open to everyone, no invitation required. You can still invite a friend from the portal as a nicety, but it is optional.

PromptParle is free. There is no paid tier and no paywall — the optimize + provider calls run on your own PC with your own keys, so there is nothing to bill. If it helps you, there is an optional pay-what-you-can way to support the project.

Open Create account, enter the code, set your password for the invited email, sign in, create a desktop license key (pp_live_…), run the installer and paste that key, then set OpenAI/Claude/Gemini/Grok keys in the local UI (⋯ → Providers) or with Set-PromptParleProviderKey.

When those options are enabled for the deployment, yes; for accounts that already completed invitation onboarding. Invitation is still required to create access.

Use Forgot password on the sign-in page. If email delivery is configured, you will get a time-limited reset link. OAuth-only accounts may not have a password until you set one in Settings.

Plans can allow concurrent desktop seats, but sharing a single login is not a substitute for proper user accounts. Prefer one person per account for auditability and key hygiene.

Bring Your Own Key. You store OpenAI, Claude, Gemini, or Grok API keys on your PC (desktop 0.25+). The desktop calls that provider directly; the provider bills your account for model usage. PromptParle does not hold those keys for day-to-day chat.

On your machine only for desktop chat: local UI ⋯ → Providers, or Set-PromptParleProviderKey (DPAPI on Windows when available). They are not uploaded to PromptParle.

Run pp, open the local browser UI, then ⋯ menu → Providers. Choose the provider, paste the key, click Save on this PC. To change a key, paste the new one and save again. PowerShell: Set-PromptParleProviderKey -Provider openai -ApiKey '…'.

A license key (starts with pp_live_) that proves this PC may use your PromptParle account. Create it under portal API Keys. The full value is shown once; the portal stores a hash. It is not an OpenAI/Claude key, set those on the PC separately.

You do, via your provider account (BYOK on the PC). PromptParle’s value is reducing how much noisy context you send. PromptParle itself is free; provider token invoices are entirely separate.

No. Provider usage is billed by the provider to your key. PromptParle is free and never sits on the token meter — there is no PromptParle fee and no hidden surcharge on any token from the model vendor.

Yes. Revoke it in the portal under API Keys. The desktop client will stop authenticating with that key; create a new one if needed.

An optional list of IPs allowed to use your desktop API keys. Empty means unrestricted. Browser portal sessions are not gated by that list the same way desktop keys are.

$0. PromptParle is free for everyone — no paid tier, no paywall, no feature locks. Optimization and provider calls run on your own PC with your own keys (BYOK), so there is no server on the prompt path and nothing to charge for. AI provider tokens are always separate (BYOK). https://promptparle.com/pricing

There is no paid plan; every feature is free. If PromptParle saves you tokens and you would like to help keep it maintained, there is an optional pay-what-you-can donation — chip in whatever it is worth to you, or nothing. No features are locked behind it. https://promptparle.com/pricing

Windows (PowerShell 5.1 or PowerShell 7+) and Linux/macOS with PowerShell 7 (pwsh) and git. The local UI runs in your browser against a server bound to 127.0.0.1 on your machine.

Open the Install page and run the Windows command (irm …/install.ps1 | iex). That bootstrap clones the project from GitHub, validates your invitation code, installs the module, and asks for your pp_live_ key.

Install git and PowerShell 7, then open the Install page and run the Linux/macOS command (curl …/install.sh | bash). Same GitHub-based flow as Windows after the bootstrap download from this site.

After install, open PowerShell and run pp. That starts the local server and opens the browser UI at http://127.0.0.1. Leave the PowerShell window open while you chat. For terminal-only chat: Start-PromptParle -Cli. Stop with the UI control or Stop-PromptParleLocalServer.

Under your user profile (for example ~/.promptparle/config.json on Unix, or %USERPROFILE%\.promptparle on Windows). That file holds the desktop license key, local provider keys (protected), and preferences. On Windows, DPAPI is used when available.

No. Remote access and repo credentials stay on your machine. They are not uploaded to PromptParle as part of normal SSH/Git/workspace use.

On desktop 0.25+, a secret gate runs on your PC before any model call (mask credential-shaped patterns; strict policy can block residual high-confidence secrets). No scanner is perfect, still avoid pasting production secrets when you can.

Prompt and context stay on your PC for optimize. Model calls go from your PC directly to OpenAI/Claude/Gemini/Grok with your local provider key. The PromptParle portal is for license and account, not on the model path. Details: https://promptparle.com/trust

The local UI can check for client updates and apply a portal-published package. You can also re-run the installer or use the module’s update command. After updating, hard-refresh the browser tab.

Yes for local work: the UI, dial optimize, and tools that only touch your machine run on your PC without sending prompts to PromptParle. Model calls need network access to your AI provider. Occasional license checks may contact PromptParle.

Account profile, hashed desktop keys, usage stats, and session titles. Prompt and context text are not stored in the cloud (stats-only product policy). Provider keys for desktop chat live on your PC, not in the portal.

No. Usage history is token stats and session titles only. Prompt and context bodies are not captured or retained in the portal.

The public site is served over HTTPS. Your connection to AI providers also uses those providers’ HTTPS APIs.

Operators who manage invitations for the deployment. Do not put secrets in the optional note field on the request form.

Client and portal source may be published under the project’s repository and license terms. That does not mean every internal system, process, or partner integration is public. Check the repository LICENSE and SECURITY docs for the authoritative statement.

No on desktop client 0.25+. Optimize and the model call run on your PC; the provider is called directly with a local key. The portal handles account, plan, and the pp_live_ desktop key. https://promptparle.com/trust

On your PC only (DPAPI on Windows when available). Set-PromptParleProviderKey or Providers in the local UI. They are never uploaded to PromptParle. The separate pp_live_ desktop key is only for license/entitlements.

Account, plan, hashed desktop keys, invitations. Not your prompt/context bodies and not your OpenAI/Claude/Gemini/Grok keys (local-first 0.25+).

Anyone who cannot send optimized context to their chosen AI provider (that path still exists, it is the model). If you need fully air-gapped inference with no external model at all, this product is not that. For hop/key-custody concerns vs PromptParle itself: local-first removes PromptParle from that path.

Often high: noisy logs, duplicated frames, fat security packs, multi-file dumps, agent chains that re-ship the same context. Often near-zero: short unique questions and already-tight prose, there is little honest bloat to remove. That is correct behavior, not a failure. See https://promptparle.com/examples

Short prompts and already-compact unique text often cannot shrink further without losing meaning. Clean prose is already signal. Savings show up more on noisy logs, duplicates, and filler-heavy packs.

No. “Keep the signal” is the goal, not a fixed percent. We are still measuring real workloads; published example packs (Noisy ~78%, Security ~60%, Clean ~2%) show the shape of results so far. Proof today is the dial savings line in the desktop UI per turn.

Open categories include: repetition/near-duplicates, boilerplate, structure-over-bulk, profile bias (security/log/dev/etc.), dial 1-5 fidelity tradeoff, and secret masking. We do not publish the full scoring IP. See https://promptparle.com/trust

In the desktop client’s savings views (per turn) and optionally in the portal Usage page (token stats and session titles). Full prompt text is not stored in the cloud.

Optimize-only paths are meant to compress and report savings without a full provider completion. Full chat/completions still call your configured model.

Common causes: code not yet accepted on the portal, typo, already redeemed, or expired. Finish the email/portal form first, then re-run the installer with the same code.

Usually a revoked, incomplete, or wrong pp_live_ key, or an IP allowlist that does not include your current address. Create a new key and check Settings → API IP allowlist.

Confirm the PowerShell window is still running, the port is free, and you are browsing http://127.0.0.1 (not a remote host). Try Stop-PromptParleLocalServer then pp again.

Use Uninstall-PromptParle from the module or the uninstall script in the repo. Optional flags can remove local config and the git clone.

Start with the desktop Help panel and the GitHub README. For account or invitation issues, reply to onboarding email or contact the operator who invited you.

Only under your organization’s policies. You must validate provider terms, data residency, logging settings, and network controls yourself. PromptParle does not replace your compliance program.