Support

Something broke. Let\'s fix it.

Something not working? Start here. Common issues, diagnostics, and honest information about what ASURIQ can and can't do.

All systems operationalEngine · Extension · API · Auth
Troubleshooting

Common issues by product

Most problems have a quick fix. Find your product below.

Extension

Check permissions: Click the extension icon in your toolbar → make sure it has permission on the current site.

Supported sites: ASURIQ works on ChatGPT (chat.openai.com), Claude (claude.ai), Gemini (gemini.google.com), DeepSeek, and other supported AI platforms. It won't activate on non-AI sites.

Try: Refresh the page. If badges still don't appear, disable and re-enable the extension in chrome://extensions.
This means the extension couldn't reach ASURIQ's verification engine. Possible causes:

• The engine may be temporarily unavailable. Wait a minute and try again.
• Your network may be blocking the connection. Check if you're behind a corporate firewall or VPN.
• Test connectivity: open a terminal and run curl https://consensus-production-6eeb.up.railway.app/health
ASURIQ supports all Chromium-based browsers: Chrome, Edge, Brave, Arc, Opera, Vivaldi, and more. Firefox is also supported via a separate add-on.

If badges work in Chrome but not another Chromium browser, check that the extension is installed and enabled in that browser's extension manager. Each browser has its own extension installation. They don't share.

Safari is not currently supported.

Routing

Format: Make sure you're using the correct key format for your provider. OpenAI keys start with sk-, Anthropic keys start with sk-ant-.

Permissions: Some provider dashboards let you restrict key permissions. ASURIQ needs chat/completions access at minimum.

Billing: If your provider account has no payment method or has hit a spending limit, API calls will fail. Check your provider's billing dashboard.
AI providers regularly update their model offerings. A model that worked last week may have been deprecated or renamed.

Check your provider's status page for current model availability. If a specific model is unavailable, ASURIQ's routing will attempt to find an equivalent alternative when possible.

Memory

Memory Local: Requires the browser extension. Data is stored in your browser's IndexedDB. If you've cleared browser data or are in incognito mode, memories won't persist.

Memory Hosted: Requires an active ASURIQ account with a Memory Hosted subscription. Check that you're logged in and your subscription is active at asuriq.dev/account.

Storage limits: Memory Local holds up to 5,000 entries. Memory Hosted holds up to 10,000.
Memory uses relevance scoring based on BM25 and cosine similarity. Older or less frequently accessed memories naturally score lower in recall rankings.

If important memories aren't surfacing, try rephrasing your query to include terms that match the original memory. Memory recall is keyword-sensitive. It matches against the content of stored entries, not semantic meaning (semantic search via ONNX embeddings is planned for a future release).

Hands

Hands Local (MCP/CLI): Tools execute on your machine. Check that you have network access, the required permissions (file system, network), and that no firewall or antivirus is blocking execution.

Hands Hosted (credits): Tools execute on ASURIQ's servers. If a server action fails, credits are automatically refunded. Check your credit balance at asuriq.dev/account.
If Claude Desktop can't connect to ASURIQ's MCP server:

1. Check your MCP configuration file. On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows: %APPDATA%\Claude\claude_desktop_config.json.

2. Restart Claude Desktop after any config changes. MCP servers are loaded at startup.

3. Make sure your ASURIQ API key is correctly set in the config.

WorkForce

WorkForce simulations run asynchronously. Your browser polls for results. If a simulation appears stuck:

• Complex deliberations (5+ models, multiple rounds) can take several minutes. The progress indicator shows which phase is active.

• If a provider times out, the simulation continues with available models and notes the missing perspective in results.

• You can close the tab and return. Simulation state is saved. Check your account for completed results.
If a WorkForce simulation fails to produce a result, credits are automatically refunded to your balance. The refund typically processes within seconds.

WorkForce uses a bid system: you see the estimated cost before the simulation runs and approve it. If the actual cost is lower than the bid, the difference is refunded. If the simulation fails entirely, the full bid is refunded.

Check your transaction history at asuriq.dev/account.

Code Quality

Code Quality requires a model capable of code analysis. If you're using Routing BYOK, make sure at least one of your configured provider keys gives access to a code-capable model (GPT-4 class or equivalent).

The quality gate system will tell you if the active model doesn't meet the minimum tier for a given analysis framework. In that case, you'll see a message explaining which tier is required and how to unlock it via Routing.

Verify

Verify cross-references claims against 18+ evidence databases. Some databases have limited coverage or may be temporarily unavailable.

If every entity in a response shows "not found," the most likely cause is a temporary issue with our database connectivity. Try again in a few minutes.

If the problem persists, note that database coverage varies by domain. Scientific claims have the broadest coverage. Niche or very recent topics may have limited reference data. This is a coverage limitation, not a bug.
Self-diagnostics

Check it yourself.

Before reaching out, run through these quick checks.

Extension version

  1. Open chrome://extensions in your browser
  2. Find ASURIQ in the list
  3. Check the version number: current release is v1.0.0
  4. If outdated, click the update button or reinstall from the store

API connectivity

  1. Open a terminal or command prompt
  2. Run: curl https://consensus-production-6eeb.up.railway.app/health
  3. You should see a JSON response with "status": "ok"
  4. If the request times out, the engine may be restarting (Railway cold starts take ~15s)

Provider status

  1. OpenAI: status.openai.com
  2. Anthropic: status.anthropic.com
  3. Google AI: cloud.google.com/status
  4. DeepSeek: Check their Discord or status page
  5. If your provider is down, ASURIQ can't process requests through that provider

Account & billing

  1. Sign in at asuriq.dev/login
  2. Check your subscription status at asuriq.dev/account
  3. Verify credit balance if using credit-based products
  4. Check that your payment method is current in Stripe
Honest disclosures

Known limitations.

We'd rather you know upfront than find out the hard way.

AI providers update their APIs without warning

When OpenAI, Anthropic, Google, or DeepSeek change their APIs, models, or rate limits, ASURIQ products that depend on those providers may be temporarily disrupted. We monitor for changes and adapt as fast as we can, but some disruption is expected. This is inherent to building on third-party AI infrastructure.

Rate limits are imposed by your AI provider, not ASURIQ

If you hit rate limits while using Routing or WorkForce, those limits come from your provider account (OpenAI, Anthropic, etc.), not from ASURIQ. We respect and relay provider rate limits. We can't override them. Upgrading your provider plan or adding additional provider keys through Routing can help.

Database coverage varies

Verify checks claims against 18+ evidence databases, but not every claim can be verified. Coverage is strongest for scientific, medical, and widely-documented factual claims. Niche topics, very recent events, and opinion-based claims may return limited or no results. A 'not found' result means we couldn't find evidence. It doesn't mean the claim is false.

Cold starts on the verification engine

ASURIQ's verification engine runs on Railway. After periods of inactivity, the first request may take 10-15 seconds as the server wakes up. Subsequent requests are fast (~2 seconds). This is a known infrastructure trade-off for a product at this stage.

WorkForce is in early access

Multi-model deliberation is computationally expensive and architecturally complex. Results depend on the models available through your provider keys. Panel diversity improves accuracy, but we can't guarantee consensus quality when only one or two models are available.

Contact

Still stuck?

We read every email. Response time is typically under 24 hours.

Important

Disclaimers

ASURIQ is a quality layer, not a guarantee

Verification badges, cognitive frameworks, and code analysis improve the quality of AI outputs. They do not guarantee correctness. AI outputs should always be verified by humans for critical decisions: medical, legal, financial, or safety-related.

Your AI provider's terms still apply

ASURIQ sits alongside your AI provider, not in place of it. Your usage of ChatGPT, Claude, Gemini, or any other provider remains subject to their terms of service, privacy policies, and usage limits.

We're in beta

ASURIQ is currently in friends-and-family beta. Features may change, and we're actively building. Your feedback directly shapes what we build next.

Your data stays yours
Prompts stay with your provider. We see analysis metadata only.
Keys never stored
One-way hash for authentication. Your credentials pass through. Never persist.
~2 second responses
Single-model cognitive tools return in about 2 seconds.
37 of 100 flagged
We ran 100 ChatGPT answers through verification. See the study →

Currently in friends-and-family beta. Built on peer-reviewed cognitive architecture.

Baars — Global Workspace TheoryACT-R — Memory Decay ModelWang et al. 2025 — Silent AgreementLi et al. EMNLP 2024 — Sparse Debate