Jump to section
Parallx Wiki
Everything you need to understand, use, and contribute to Parallx — feature guides, release notes, and community handbook.
What is Parallx
An open AI model marketplace built around community and resilience.
Parallx is a platform where independent developers publish their own AI model endpoints and users access them through a single, unified chat interface. Instead of being locked into one provider, you get access to a diverse ecosystem of models — each with its own strengths — with automatic fallback routing so the conversation keeps going even when individual models go down.
For Users
- ›Chat with any published model
- ›Automatic fallback if a model is down
- ›Full session history
- ›Rate responses to improve models
- ›Community model pool access
For Developers
- ›Publish your own AI endpoints
- ›Per-model usage analytics
- ›Feedback collected as training data
- ›Smart fallback chain handles outages
For Contributors
- ›Donate free-tier API keys
- ›Join the community model pool
- ›Help keep the platform always-on
- ›Earn a Contributor badge
Release History
What's been added and improved in each version, newest first.
The platform's most significant update: anyone can now register a community model and contribute API keys to keep it running. Multiple contributors can back the same model, forming a shared pool that the Smart Router draws from automatically when capacity is needed.
Community model submissions
Any logged-in user can submit a model for the marketplace. No developer role required.
Shared key pool
Multiple contributors can donate API keys to the same model. Keys rotate automatically on quota exhaustion or failure.
Smart deduplication
Submitting a model that already exists on the platform adds your key to the existing model's pool rather than creating a duplicate.
Admin review
All community models and new key contributions go through admin review before going live. Admins can test a key before approving it.
Two-phase routing
The Smart Router tries the same model's pool first, then any other community model, before escalating to developer-hosted models.
Automatic daily reset
Models that are auto-disabled due to repeated failures are automatically re-enabled each night so they get a fresh chance.
Issue streak tracking
Admins can see how many consecutive nights a model has been auto-disabled — useful for identifying chronically failing models.
Removal advisories
When an admin removes an approved community model, a notice is broadcast to all users so nobody is caught off guard.
Parallx is now properly discoverable on Google, Bing, and AI-powered search tools like Perplexity and ChatGPT web search.
- Rich preview images when sharing links on social media.
- Each published model appears in search engine sitemaps.
- FAQ-style rich results eligible to appear directly in Google search pages.
- AI search crawlers can now accurately describe what Parallx does.
Parallx is now installable as a home screen app on Android and iOS — no app store required.
- Tap “Add to Home Screen” in your browser to install instantly.
- Full-screen app experience: no browser chrome, violet theme, branded icon.
- Works offline for previously visited pages; syncs when back online.
Found a bug or have a feature idea? You can now report it directly from any page on Parallx — no external tools needed.
- Floating “Report an issue” button on every user-facing page.
- Attach up to 3 screenshots per report.
- Reports include the page you were on automatically.
- Track the status of your reports (Open → In Progress → Resolved).
Your feedback now means more — and you can see exactly what it does.
- After rating a response, the widget shows the model's live average rating and your total contribution count.
- A tooltip explains how your ratings are used to improve models over time.
- Developers now see token usage and response-completion reasons in their analytics.
- Chat sidebar: quickly jump between recent sessions and switch models without leaving the chat.
- Inline session rename — hover any session in the sidebar to give it a custom name.
- Retry button: when a message fails, one click removes the error and resends automatically.
- Bonus sessions awarded for rating messages and end-of-chat feedback.
- New upgrade page with Free vs Pro feature comparison.
- Responses now stream word-by-word as the model generates them — no more waiting for the full reply.
- Markdown renders progressively: code blocks, tables, bold, lists all appear as they arrive.
- Smarter fallback: the router now retries automatically before surfacing an error, making outages nearly invisible.
- Session history page with the ability to resume past conversations.
- Free-tier session cap (10 sessions); earn more by rating responses.
- Marketplace homepage with model grid, categories, and featured models.
- Chat interface supporting OpenAI-compatible, Gemini, and custom model formats.
- Smart Router with automatic fallback — if one model fails, another takes over seamlessly.
- Feedback system: star ratings per message and session-level rating collected as training data.
- Developer portal: publish models, view analytics, receive notifications.
- File upload: attach PDFs to any conversation and the model reads them.
- Terms of Service covering data use and feedback for model improvement.
Community Pool Guide
How to submit a model, contribute keys, and understand how the pool works.
The Community Poollets any Parallx user register an AI model backed by their own provider API key. When multiple users contribute keys to the same model, those keys form a shared pool that rotates automatically — so the model keeps serving requests even when one person's quota runs dry.
How routing works
When you send a message, the Smart Router works through a chain of options in order until it finds one that works:
Submitting a community model
Go to the Community Hub
Navigate to Community Models in the navigation and click "+ Submit Model". Any logged-in user can submit — no special role required.
Fill in the details
Give your model a name and description, pick a category, choose the API format (OpenAI-compatible or Google Gemini), enter the model identifier and the endpoint URL, then paste your API key.
Automatic deduplication
If a model with the same endpoint and model identifier already exists and has been approved, your API key will simply be added to that model's key pool. You'll see a confirmation screen explaining this — no duplicate is created.
Wait for admin review
New models and new key contributions are reviewed by an admin before going live. Admins test the key directly to confirm it works. You'll see a Pending Review status badge on your submission in the Community Hub.
Manage your keys
Once approved, visit your model's Manage page from the Community Hub to add more keys, pause individual keys, or check their status.
Key status explained
| Status | What it means | What to do |
|---|---|---|
| Pending Review | Submitted, awaiting admin approval. | Nothing — the admin will review shortly. |
| Approved | Active and serving requests from the pool. | Nothing — your key is working. |
| Out of pool | You manually paused this key. | Toggle it back on from the Manage page. |
| Auto-disabled | The model hit repeated failures and was paused by the system. | Check the model's health; it resets automatically at midnight UTC. |
Nightly reset
Every night at midnight UTC, any community model that was automatically disabled during the day is re-enabled and given a fresh chance. If a model keeps getting disabled night after night, admins can see its consecutive failure streak and decide whether to intervene.
Security & privacy
Can other users see my API key?
No. Your API key is stored securely on the server and is never exposed to any other user, contributor, or browser — including admins.
Can I see the messages my key processes?
No. You only see aggregate stats like request counts and key status. No message content is ever visible to contributors.
Can I remove my key at any time?
Yes. Go to Community Hub → your model → Manage, and remove any key. It stops being used immediately.
Does contributing a key violate my provider's terms?
Review your provider's free-tier policy. Most personal API keys allow this kind of use, but you are responsible for ensuring compliance with their terms.
What if I think my key was compromised?
Revoke it immediately from your provider's dashboard, then remove it from Parallx via the Manage page. The platform will automatically stop using it on the next failure anyway.
Version History
Archived documents from earlier iterations of this wiki.
The original wiki described the Community Pool as a proposed feature. It shipped in v0.8.0 and is now fully documented in the Community Pool Guide above. The archived version is preserved below for historical reference.
Show archived content ↓Hide archived content ↑
The 30-second version
Most AI providers give you a free daily quota that resets at midnight. You probably don't use all of it. Instead of letting those unused tokens expire, you could donate them to the Parallx community pool — and in return, get access to what everyone else has contributed. Like a neighborhood tool library, but for AI compute.
Share what you don't use. Access what others share. Everyone wins.
Why it was proposed
The platform's Smart Router already handled developer model fallbacks, but had no safety net when all developer-hosted models were unavailable. The proposal was to add a community layer below the developer chain — a shared pool of free-tier keys contributed by ordinary users, rotating automatically on failure or quota exhaustion. The goal was zero-cost resilience: keeping the platform available even during outages, without requiring paid infrastructure.
Original design principles
- Rotate on quota exhaustion (429) and server errors — not on invalid key errors (those keys are simply skipped).
- Any HTTPS endpoint accepted, but every model stays inactive until an admin reviews and approves it.
- Per-user moderation: admins can restrict users who repeatedly submit failing models from creating more.
- Supported formats: OpenAI-compatible and Google Gemini only — no custom formats in the pool.
What evolved during build
- Multiple keys per model were added — the original design assumed one key per model.
- Deduplication was built in: same endpoint and model name merges into the existing model rather than creating a duplicate entry.
- The duplicate submission flow became a pre-filled form with locked fields, giving contributors a clear path to join an existing pool.
- Admin key review was separated from model review — pending key contributions from duplicate submissions get their own review queue.
- Issue streak tracking and a nightly auto-reset were added for operational visibility and daily recovery.
- An advisory notification system was built so users are informed when an approved community model is removed by an admin.
Parallx Platform Team · Updated April 20, 2026
← Back to marketplace