Classify
Map the request to a known task, required modalities, context size, tools, and risk.
AI ROUTING CONTROL PLANE
One request. The right AI for every step.
Tell AI Hub what done looks like. It maps the work, checks your policies and connected accounts, then recommends the strongest route for your priorities before any provider sees the request.
APP BUILDER
Turn an idea into an app.
Save projects, preview changes, run tests, export code, and publish for the web or build Android packages. Use your own connected AI account.
Provider-directNo intermediary model gateway
Cost-awareExact usage when providers report it
Policy-boundPrivacy and spend gates first
Receipt-backedAuditable decisions and execution
YOUR ROUTE
Start with an outcome
Your route will appear here with a plain-language recommendation, capability fit, provider state, exact or estimated cost labeling, and eligible alternatives.
DURABLE EXECUTION / LOCAL
Managed workflow
Approve, execute, resume, and audit a task graph without storing prompts, evidence, or model outputs in the workflow record.
Checking workflow capability...
CONSUMER EXECUTION / OFFICIAL CLI
Run through your provider plan
Choose Auto route for a no-provider-call analysis and explicit approval, or select a specific signed provider connector. API keys are never used by this console.
Sign in and connect Codex or Claude Code to begin.
MULTI-STEP ORCHESTRATION / CONSUMER PLANS
Route the work, not just the prompt.
AI Hub decomposes compound work locally, chooses a policy-compliant provider for every step, shows the complete graph, then waits for one-time approval before any provider is called.
Checking tenant workflow policy and signed connectors...
MODEL ATLAS
A living map, not a leaderboard.
Documented capabilities and prices establish eligibility. Provisional task scores establish the first route. Evaluations will replace opinion with customer-specific evidence.
CONNECTION VAULT
Your providers. One control plane.
Choose an encrypted API connection or pair an official consumer client. Consumer passwords, cookies, and provider tokens never enter AI Hub.
NORMALIZED LEDGER
Every run. One economic truth.
AI Hub reconciles direct API and official consumer-client runs without pretending an unreported consumer cost is zero. Prompts, outputs, and credentials are deliberately excluded.
Sign in to view private usage.
EVIDENCE-DRIVEN ROUTING
Teach the router from verified outcomes.
Feedback binds to a reconciled execution receipt. It never stores prompts, outputs, or free-form customer text.
Sign in to inspect outcome evidence.
A completed, unevaluated execution and a 1–5 rating are required.
CONTROLLED EXPLORATION
Challenge a proven route safely.
AI Hub selects one already eligible route. It never calls both providers merely to compare their answers.
Sign in to inspect route experiments.
Sign in first.
ONE ACCOUNT / EVERY AI HUB APP
Your account. Your devices. Your control.
Use one AI Hub username and password on Android, Windows, and the web. This account identifies your private workspace; OpenAI, Anthropic, and other provider credentials stay in their official clients or the encrypted API broker.
Sign in before managing security
Account security, device sessions, recovery, provider connections, and routing history are private to your AI Hub workspace.
Private workspace
Plan and billing
Choose a plan here, pay on Stripe's hosted checkout, and return with access confirmed directly from Stripe. AI Hub never receives or stores card numbers.
No refunds. AI Hub is a paid product, not a donation. Purchases and renewals are non-refundable except where required by law. No-refunds and cancellation policy.
Checking billing access...
Sign in as an organization administrator to manage billing.
Stripe handles card entry, authentication, receipts, and payment-method storage. AI Hub activates service only after a signed webhook or server-to-Stripe confirmation.Organization members
Invite teammates into this private workspace. Active members and pending invitations count against the plan's seat limit.
Share this code securely. It is shown once.
Members
Invitation history
PLATFORM COMMERCIAL ADMIN
Revenue operations
See conversion and recurring-revenue signals without exposing customer identities.
Sessions you control
Session tokens are never displayed. The server stores only their hashes.
Recent account activity
Bounded metadata only. Passwords, bearer tokens, provider credentials, prompts, and outputs never appear here.
Protect access and recovery
This account uses a password, one-time recovery codes, session controls and sign-in rate limits. Save recovery codes privately and use a unique password. Local AI Hub accounts do not currently support passkeys, email-based recovery or multi-factor authentication; do not treat them as if those safeguards were enabled. For access or security concerns, contact support@autosecurelogin.com. Review the account and subscription guide before connecting sensitive work.
DIRECT PROVIDER CONTROL / ENCRYPTED
Put every API account to work.
Connect multiple first-party provider accounts, assign budgets and routing priority, and see when health protection removes an account from service. AI Hub resolves keys internally and never returns them to this page.
Sign in to inspect provider accounts.
ROUTING READINESS
Encrypted account inventory
Enter a change reason above before enabling or disabling an account. Verification never sends a model prompt.
LIVE AI COORDINATION / PLATFORM ADMIN
Every task. One shared picture.
See what each AI session is doing across your computers, pass work between sessions, and keep one shared memory document current. Identity comes from registered workstation keys, not self-reported machine names.
Checking task coordination access…
WORK QUEUE
Assign, claim, and finish real work
Dependencies must finish before a session can claim downstream work. Every update uses a version check, so one computer cannot silently overwrite another.
SESSIONS
Who is working now
Online status is a short server lease. A stale task is shown honestly rather than guessed to be running.
No registered task sessions have reported yet.
SHARED MEMORY
One durable context file
All registered sessions receive the current version when they report. Conflicting edits stop safely so newer context is never silently overwritten.
ORGANIZATION CONTROL / TENANT
Policy with teeth.
Control which providers, tasks, modes, budgets, approvals, and retention limits the routing engine may use. The server enforces these settings; clients cannot loosen them.
Sign in to inspect organization policy.
POLICY AUDIT
Recent revisions
Actor identifiers are hashed; policy content and credentials are not written to this audit view.
Audit permission required.
HOW ROUTING WORKS
Capability first. Evidence next. Economics last.
A cheap wrong answer is still expensive. AI Hub starts with hard capability gates, then optimizes the models that remain.
Gate
Remove models that cannot satisfy the task, policy, lifecycle, or known budget.
Score
Blend task quality, speed, estimated cost, credit headroom, and stability using the selected mode.
Explain
Show the preferred route separately from the best route the customer can execute now.
Execute
Call the provider server-side, capture usage and request IDs, and fail over only on transient failures.
Learn
Measure outcomes by task family and replace provisional scores with reproducible evaluations.
START HERE
From account to approved route
AI Hub is the control plane between your work and the AI providers you choose. It understands the task locally, checks connections, policy, budget, and privacy, then shows the route before anything is sent.
- Create or sign in to AI Hub.Your account keeps organization policy, projects, provider connections, usage, approvals, and receipts together.
- Open Connections.Add an encrypted API account for automated work, or connect a supported consumer-plan companion for approved desktop execution.
- Describe the completed outcome.Include context, constraints, required format, deadline, and acceptance criteria. Compound work can become separate steps.
- Choose a routing mode.Quality, Balanced, Economy, Fast, and Conserve change ranking priorities, never security or capability requirements.
- Review the proposed route.Check the model, alternatives, cost status, required connectors, risks, and approval gates.
- Approve execution.Provider calls occur only after required approval. Completed runs return reported usage and an auditable receipt.
CONNECTIONS
API accounts and consumer plans are different
- API account
- An organization-owned provider credential used by the broker. The browser never retrieves it; the broker returns output, usage, cost, and a receipt.
- Consumer plan
- A supported official desktop or command-line session such as ChatGPT/Codex or Claude. It is separate from API billing and needs the local companion installed, open, signed in, and approved.
Signing into a provider website does not create an API connection. Each provider card states which connection type is available.
TASK REQUIREMENTS
Some outcomes require more than a model
- Drive, Notion, email, calendar, CRM, and similar searches require an approved connector with the necessary scope.
- Image, video, voice, OCR, document parsing, and code execution require a model and runtime that explicitly support that capability.
- Confidential or regulated work can require a private route, a specific provider account, retention controls, or an additional reviewer.
- If something is missing, AI Hub blocks the route and names the connector or capability that must be enabled.
ROUTING MODES
Choose what matters for this request
- Quality
- Favors the strongest eligible model within policy.
- Balanced
- Combines quality, cost, speed, reliability, and remaining headroom. This is the daily default.
- Economy
- Favors lower-priced eligible routes while preserving minimum quality.
- Fast
- Favors lower latency and fewer handoffs when possible.
- Conserve
- Protects scarce credits and shifts work toward connected alternatives with headroom.
COST & USAGE
Know what the number means
- Exact: reconciled from usage returned by the provider.
- Estimate: calculated before execution from catalog pricing and expected units.
- Unknown: not enough billing data was returned. AI Hub does not present zero as proof that work was free.
- Provider credits: promotional or prepaid balances remain provider-controlled and may differ from local headroom.
PROJECTS & POLICY
Keep customer work isolated
A project can bind approved accounts, connectors, budgets, privacy defaults, members, and run history. Tenant-wide routing is useful for catalog preview, but project work should select a project before planning or execution.
Request controls can tighten organization policy; they cannot weaken it.
BUTTONS & STATUS
No unexplained dead ends
- A disabled action must explain whether sign-in, a request, a connection, a policy decision, or administrator setup is required.
- Analysis progress and the completed result are moved into view so the user can see that something happened.
- Orange needs attention and green is ready or accepted, but the nearby text is always the authoritative explanation.
COMMON PROBLEMS
What to do when a route cannot run
Sign-in required: Open My account and sign in, then return to Connections. Local preview can show the catalog without enabling customer connections.
Companion unavailable: Install it in the standard location, open it, sign into the intended provider, and retry detection. Website login alone is not enough.
Connector required: Open Connections and approve the named connector. AI Hub will not replace a business-data search with an ordinary web answer.
No eligible model: Review capability, privacy, region, budget, and account requirements. Connect another provider or adjust only allowed constraints.
Cost unavailable: Treat the amount as unknown until provider usage is returned. The provider dashboard is authoritative for balance and billing.
Connection failed: Confirm the local service address, network reachability, account status, and provider authorization before retrying.
READABLE TASKS & WORKER STATUS
Know what each computer is actually working on
Before submitting a consumer task, enter a title, project display label, purpose, stage and expected result. You can also name the larger task. For example: “Render forest scene frames 120–239”, project “Forest cartoon”, purpose “Prepare the forest scene for episode one”, stage “Rendering”, result “120 image frames for assembly”. These labels are shared with the assigned worker and saved with the task, so omit confidential prompts, passwords and file paths.
- For a single task, fill in the description and private request separately. Automatic routing shows the saved description before you approve the chosen provider.
- For a workflow, analyze the graph first. Describe what each step will accomplish, then choose Review work descriptions. This saves the descriptions without executing anything.
- Review those saved descriptions and select Approve and start workflow. Editing a description requires another review; the previous approval token cannot authorize the changed plan.
- Companion 0.12.0 shows the approved task title while it runs, with purpose, project label, stage and expected result below it. Status changes such as running or completed are separate from the producer's stage label.
Description unavailable: the producer did not supply the new description fields. It does not mean the worker is idle. Upgrade that producer; do not guess the work from a session ID or reveal its private prompt to create a label.
Description save not confirmed: execution remains paused because storage could not confirm which approval is current. Have the operator inspect the recorded storage error and perform controlled recovery. Do not repeatedly approve or restart shared services blindly.
The project display label grants no access: authenticated tenant identity, immutable project policy, connector ownership and local resource admission remain separate authorities. This consumer-task feature alone is not GPU distribution or file transfer; those four-product integration capabilities have separate release and acceptance gates.
Support: contact@autosecurelogin.com. Include the app version, time and error code. Never send provider credentials, pairing tokens or private task content.
SECURITY BOUNDARIES
What AI Hub does and does not do
Task classification and route planning happen before provider execution. Credentials remain in the encrypted provider broker and are not returned to the PWA. AI Hub uses direct provider integrations rather than an intermediary model gateway, enforces tenant and project boundaries, and records approvals and receipts.
A recommendation is not execution. A catalog listing is not proof that an account is connected. An estimate is not final billing. A provider call is not complete until usage and receipt status are reconciled.