Model Hub

• Providers & routing configures where models come from and how each Agent reaches them.
• Tools shows the tools Vibey can call, including web search and local-first web fetch, and what real use showed about them. See Use hosted tools.
• Usage shows calls metered by Model Hub. See Explore usage.
• Subscription quota shows the limits of each subscription added as a Model Hub provider and what its use is worth at API prices. See Check subscription quota.

• Providers lists the subscriptions, model API keys, and search/fetch services you own. It is an inventory, not a priority list; a service-only provider supplies no models.
• Routes shows each Agent backend and the effective Route chain for each model, inherited from default routing or saved as a manual override.

Choose who manages a backend

Migrate saved API keys

Add a provider

Add a subscription

1. In Providers, select Add subscription and choose the provider.
2. For Claude, choose who holds the sign-in. Add as a Model Hub provider is selected by default; Use own login is the other option.
3. Select Sign in and complete the provider’s sign-in flow.

• Claude: with Use own login, Claude Code keeps using its local login instead, and that subscription supplies only Claude Code. Anthropic’s subscription terms cover only its own apps, such as Claude Code and claude.ai; Avibe shows this as one line under the Claude option. Read Anthropic’s terms before you add a Claude subscription to Model Hub.
• ChatGPT and the other subscriptions are added to Model Hub only. Native Codex sign-in remains supported when Codex is already signed in, but quota exhaustion will not trigger Model Hub takeover and its usage stays outside Model Hub.

Add an API key

1. Select Add API key.
2. Choose a Provider. Picking DeepSeek, Qwen, Kimi, and the like fills in the official address and API format; for anything else, choose Other provider · enter its address. Then enter an optional name, the Base URL, and the API key. A custom Base URL can point to a relay, aggregator, or self-hosted compatible service.
3. Select Add. Avibe connects once, authenticates, identifies the interface, and fetches the model list before saving.

Maintain the model list

• Select Refetch to test the saved provider, update its health, and refresh discovered models. This is the only saved-provider refresh action.
• Select Add model ID to add a model manually when discovery cannot find it.
• Edit the exact reasoning tiers accepted by any discovered or manual model. No tier is selected or filled by default.
• Remove models you added manually. Models reported by the provider are managed by refetch instead.

Configure default routing

Manage a backend’s model list

Add a model

• A catalog model that no provider lists appears under Model catalog, marked Not listed by a provider. Selecting it adds the model with the catalog’s details: the vendor’s own model ID (for example claude-fable-5-2), context and output limits, input types, and reasoning efforts. No form opens. Avibe does not check that one of your providers serves it; if none does, the chat says so when the model is used.
• If neither your providers nor the catalog has the model, Add “x” as a custom model adds exactly the ID you typed, with default settings and no form.
• Add custom model… opens the full editor, for an ID with settings of your own.

Add from a model picker

• Add model appears only for a backend managed by Model Hub, and only for the instance Owner, who is the only person who can change Model Hub. A backend on its own login lists its own native models.
• If adding fails, a notice says so and the dialog keeps your choices on screen.

Use fast mode

• Fast is offered when Model Hub serves the model through an API key or subscription that speaks the model’s own protocol: an OpenAI or Anthropic API key, on the official endpoint or a relay, or a ChatGPT or Claude subscription that Model Hub holds. It is not offered for an Agent on its own login or for OpenCode.
• Each reply reports whether it actually ran fast. A relay may drop the request or serve standard speed. When a fast reply is not confirmed, the bolt next to the menu’s arrow turns hollow, and a note under the reasoning effort says so. Fast stays on, and you can still turn it on or off. The next fast reply is checked again, and the bolt fills in once the provider confirms it. Usage counts only confirmed fast replies at the fast price.
• If you switch to a model without a fast tier, your choice is kept but the chat runs at standard speed. When you switch back, it runs fast again.
• A new chat in the Web interface starts with your last choice. Chats started from a chat app, scheduled tasks, and vibe agent run use standard speed.
• If the provider refuses fast mode, for example on a Claude subscription without usage credits, the reply runs at standard speed instead of failing.
• Fast mode costs more. OpenAI and Anthropic bill it at their own higher rates, a ChatGPT plan uses more of its quota, and a Claude plan pays for it from usage credits. Check your provider’s pricing.

Edit a model route

• add or remove a hop;
• reorder hops;
• change a hop’s explicit upstream model mapping, including an exact ID absent from an API-key provider’s model list; and
• select Save to store a nonempty chain as a manual override.

Restore automatic routing

Read the status

Explore usage

Search and fetch service usage

Check subscription quota

• Last 7 days at API price covers the trailing seven days.
• This cycle at API price covers the current billing cycle when the account has a renewal day set in the price override file. Otherwise the page shows Last 30 days at API price.
• Paid back compares that period’s value with the plan’s monthly fee, as a multiple of the fee or as the amount still to pay back.

Correct prices

{
  "models": {
    "my-relay-model": {"input": 3, "output": 15, "cache_read": 0.3, "cache_write": 3.75},
    "relay-alias": {"alias": "claude-sonnet-5"}
  },
  "plans": {"team_seat": {"fee_usd": 30}},
  "sources": {"src_abc123": {"plan": "team_seat", "renewal_day": 14}}
}

• models is keyed by model ID. Prices are USD per 1 million tokens. input and output are required; cache_read, cache_write (five-minute cache), and cache_write_1h are optional. A missing cache_read or cache_write is priced as input, and a missing cache_write_1h as twice input. Instead of prices, an entry can give alias to use another model’s price. Model IDs are matched without a date suffix, a bracketed suffix such as [1m], or a provider/ prefix. Keys longer than 128 characters are ignored.
• plans sets or adds a monthly fee in USD for a plan key. The built-in keys are claude_pro, claude_max_5x, claude_max_20x, chatgpt_plus, and chatgpt_pro.
• sources is keyed by provider ID, as shown in the Provider ID column of the Usage CSV export. Provider IDs that Avibe creates look like src_abc123. If that column shows a shortened internal ledger key instead, the override cannot match it. plan sets the account’s plan and wins over the plan the vendor reports. It can be a built-in key, a key you added under plans, or a vendor plan name such as max 20x. renewal_day (1–31) sets the day of the month the billing cycle renews; a day past the end of a short month renews on its last day.

Use hosted tools

Add a search service

1. Open Models > Tools > Web search > Add search service, or Providers > Add provider > Search services.
2. Choose Tavily, keyed Exa, or DeepSeek. The dialog shows each vendor’s price, free-tier information, and facts about where queries go.
3. Enter the vendor’s API key and save. Tavily and keyed Exa create service-only providers for search and fetch, with no models. DeepSeek creates an ordinary DeepSeek provider for models, search, and fetch, using the official endpoint; an existing qualifying provider reuses its key. A DeepSeek relay does not inherit these service APIs.

Built-in keyless Exa

How a search runs

• Native: the model’s vendor searches within the model call.
• Search service: Model Hub sends the query to Tavily, keyed Exa, the built-in keyless Exa service, or DeepSeek’s search API. The built-in Exa service does not use a key.
• Model executor: another provider’s lightweight model runs its vendor’s native search on Vibey’s behalf. The provider must list an eligible executor model.

Google Search on Gemini

• Where the query goes: Google runs the search inside the model call. Its Gemini API Additional Terms say Google stores the prompt, any context you provide, and the output for 30 days to create grounded results.
• Cost: Google bills each search query the model runs. Gemini 3 and later models share 5,000 free queries a month, then pay $14 per 1,000 (Google’s price, reviewed October 29, 2026). One answer can run several queries. Google’s default limit on the paid tier is 1,500 grounded requests a day per project.
• Sources: Google returns each source as its own redirect link on vertexaisearch.cloud.google.com, so a citation’s link text shows that domain. When the model searches and then calls another tool, the sources appear as citations on the answer, not on the search card.
• Search Suggestions: Avibe does not show the Search Suggestions Google returns with grounded results, although Google’s terms ask applications to show them. Review those terms for your use. To keep Gemini on Search order, turn off Prefer the model’s own search, which applies to every model.

Read a web page

• The tool switch disables web fetch. Each row has a separate Use for web fetch switch; turning off a service’s fetch switch leaves search unchanged.
• Fetch order is always visible. Reorder with the up/down buttons; disabled rows remain visible at the end. Reset to automatic removes your manual order. Jina is last by default, not fixed there after reordering.
• Changes take effect next turn. When a page is unreadable or an executor is unavailable, the chain tries the next eligible entry, except for terminal policy refusals or a local login wall. Retry now clears a service cooldown; it does not make a test request. Exa’s cooldown is shared with web search, although its two switches are independent.

Where the URL and content go

Fetch costs and limits

Compatible endpoints

• On an OpenAI Responses endpoint, gpt- models run OpenAI’s web search, and grok-4 models run xAI’s web search and X search. It can also run proxied searches on those vendors’ lightweight models, and offer gpt-image-2, if it lists them.
• On an Anthropic Messages endpoint, Claude models run Claude’s web search, and a listed Claude Haiku model can run proxied searches.
• On a Chat Completions endpoint, no search runs natively, and Gemini’s image models are offered if the endpoint lists them.

Status and refusals

Recover a search service

Generate images

• the vendor’s list price where there is one, such as $0.02 / image for Grok Imagine;
• an estimate from token pricing, marked ≈, such as ≈ $0.067 / image for Nano Banana;
• No unit price where neither is known, such as gpt-image-2, which OpenAI prices per image token;
• Included in subscription for a model on a subscription provider.

Costs

• An API key is valued at the vendor’s list price. A relay may bill differently.
• A subscription adds no charge.
• Native and model-executed searches follow the vendor’s search and model-token pricing. Search-service requests are priced separately; any model tokens used to read their results still belong to the model call.
• A Claude subscription does not supply hosted search or image generation; web fetch uses its own local/service chain.

On a provider

In a conversation

Before downgrading

Understand errors and recover

Latest recorded turn

Messages on the Models page

Messages in a turn

Recovery by cause

• Cooldown or takeover: wait for the displayed retry time. Recovery and the return to the first hop are automatic and do not change the chain.
• Expired subscription login: follow Sign in again for that provider.
• Exhausted balance: top up with the vendor, then open provider details and select Refetch.
• Revoked API key: replace the key in the existing provider, then refetch. The provider identity and every Route position stay intact.
• Restricted account: follow the vendor remedy or move the affected model to another provider.
• Unclassified Error: fix the suspected upstream problem, then use Refetch. A successful current check clears the blocker.
• Model rejected by the upstream: verify the exact model ID and upstream support, then edit the model or route. Absence from an API-key provider’s fetched list alone does not mean the model is unusable; subscriptions retain their existing model checks.
• Provider missing: edit the Route chain. Adding a new provider does not repair a hop that refers to the removed provider.
• Native CLI unavailable: restore the backend’s local CLI and login, then retry.
• Need an immediate escape hatch: select Use own login for only the affected backend. Your providers and Route chains remain saved for when you select Let Model Hub manage again. Vibey has no own-login option; edit its model’s Route chain instead.