Frequently asked questions

Token Harbor is a unified AI gateway that gives developers and teams access to frontier AI models through a single API and dashboard.

Rather than maintaining the largest possible model catalog, Token Harbor focuses on curated frontier models from leading AI providers, helping users access the latest and highest-performing models without managing multiple accounts, API keys, billing systems, or provider integrations.

Token Harbor provides:

• frontier AI model access
• TH Orchestra intelligent orchestration
• free model access programs
• usage tracking
• unified billing
• provider failover
• API key management
• AI agent integrations

Users can either select a specific model directly or use TH Orchestra, which automatically classifies and routes tasks through specialized model pools for planning, coding, reviewing, and summarization.

We also prioritize trusted infrastructure and reliable providers. Whenever possible, Token Harbor works with established AI and cloud platforms to deliver stable, secure, and high-availability access to frontier models.

Token Harbor is designed for:

• Developers building AI applications
• Startups using multiple AI models
• AI agents and automation workflows
• Teams managing AI usage across projects
• Non-technical users who want easier AI access

Managing multiple AI providers separately becomes complicated as your usage grows.

Token Harbor simplifies:

• access to frontier AI models through a single API
• unified billing and wallet management
• model discovery and comparison
• provider failover and reliability
• usage monitoring and analytics
• AI agent and workflow integrations

You can access models from multiple providers without managing separate accounts, API keys, payment methods, or integrations.

For users who prefer intelligent orchestration, TH Orchestra can automatically coordinate planning, coding, reviewing, and summarization workflows across specialized model pools.

We focus heavily on:

Transparency — users should understand which model handled a request, which provider served it, how costs are calculated, and how billing is tracked. We believe AI infrastructure should be observable, predictable, and transparent.

Direct Model Access — when you select a model, your request is sent directly to that model. No hidden model switching, no surprise substitutions. What you choose is what you use.

TH Orchestra — for users who prefer intelligent orchestration, it automatically classifies tasks and routes them through specialized model pools for planning, coding, reviewing, and summarization. Users can choose between direct model access and orchestration depending on their workflow.

Frontier Models — we focus on a curated set of frontier AI models rather than maintaining the largest possible catalog. Our goal is to provide access to the latest and highest-performing models from leading AI providers.

Trusted Infrastructure — Token Harbor prioritizes reliability and long-term stability. We work with established AI and cloud providers whenever possible and maintain provider redundancy for many popular models. If one provider experiences issues, requests can automatically fail over to another provider serving the same model.

Built for Agents and Developers — Token Harbor is designed to work seamlessly with Claude Code, OpenHands, AutoGen, CrewAI, LangChain, and custom AI agents, as well as traditional AI applications and developer workflows.

Token Harbor focuses on frontier AI models from leading providers. Our catalog is intentionally curated to include the latest and highest-performing models rather than every model available on the market.

Current providers include:

• Anthropic Claude
• OpenAI GPT
• Google Gemini
• DeepSeek
• Qwen

New frontier models are added regularly as they become available. The latest supported models, pricing, context limits, and capabilities are always available on the Models page.

Some models may also be offered for free as part of promotional programs. Free model availability and usage limits are shown on the Models page.

TH Orchestra is Token Harbor's orchestration model. Instead of sending every request to a single fixed model, Orchestra analyzes the task and routes it through specialized model pools.

Examples include:

• Planning and task design
• Coding and implementation
• Review and verification
• Context summarization

This allows complex workflows to use the most suitable model for each stage without requiring users to manually switch models.

When you select a model directly, every request is sent to that specific model — Claude → Claude, Gemini → Gemini, GPT → GPT.

When you select TH Orchestra, Token Harbor automatically classifies the task and chooses the most appropriate model pool for that stage of work.

Orchestra is designed primarily for coding agents, automation workflows, and multi-step tasks.

Token Harbor continuously monitors provider health and availability through internal health checks. If one provider becomes unavailable, Token Harbor can automatically route requests to another available provider serving the same model.

This helps improve:

• uptime
• reliability
• request success rates

In most cases, users continue using the same model without needing to take any action. For TH Orchestra, provider selection and failover are handled automatically as part of the orchestration process.

Yes. Token Harbor may provide free access to selected frontier models as part of promotional programs and community initiatives.

Currently, DeepSeek V4 Flash is available as a free model.

Free model availability may change over time as new models are added and promotional programs evolve. Usage limits and eligibility requirements depend on account status and are described in the Billing & Usage section.

Yes. Token Harbor uses an OpenAI-compatible API format, making migration simple for most applications.

Example:

from openai import OpenAI

client = OpenAI(
    base_url="https://tokenharbor.ai/v1",
    api_key="YOUR_API_KEY"
)

Yes. Streaming is supported using OpenAI-compatible APIs.

Yes. Token Harbor is designed for AI agent workflows and multi-model orchestration.

Compatible tools may include:

• Claude Code
• OpenHands
• LangChain
• AutoGen
• CrewAI
• OpenDevin
• custom agent frameworks

Yes. Users can create multiple API keys for separate projects, environments, AI agents, team usage, and different spending policies.

Each key may eventually support independent quotas and provider restrictions.

Planned / partial support may include:

• embeddings
• image input
• vision models
• PDF processing
• audio models

Availability depends on upstream provider support.

Token Harbor uses pure pay-per-token pricing. Every model listed on the /models page includes input pricing, output pricing, context limits, and provider information. Pricing is shown in USD per 1M tokens.

There are no subscriptions, no monthly minimums, and no required commitments. You simply top up your wallet balance, and API usage is deducted automatically based on actual token consumption.

TH Orchestra does not charge a separate orchestration fee. You are billed based on the actual model that processes your request, using the same per-token pricing shown on the Models page.

Volume discounts apply in the same way as direct model usage. The upstream model used and the final request cost are visible in the dashboard.

The dashboard is divided into two main sections.

Usage — a complete view of your account activity, including model usage, usage trends, request history and logs, request costs, and status and cache information. This lets you monitor both overall usage and individual request activity in one place.

Billing — your wallet and payment activity, including current balance, top-up history, spending history, cashback rewards, referral rewards, and promotional credits. You can also manage wallet top-ups and review your complete billing history.

Token Harbor uses a prepaid wallet system. When requests are processed:

1. 1.token usage is calculated
2. 2.pricing is determined based on the routed model/provider
3. 3.usage costs are deducted from your wallet balance

Balances update automatically after each request.

Yes. Users can configure daily spend limits, monthly budgets, per-key quotas, and provider allow/block lists.

These controls are especially useful for teams, AI agents, automation systems, and shared environments.

Yes — unused wallet balance is refundable. To request a refund, email billing@tokenharbor.ai within 30 days of the original top-up.

Refund policy:

• unused balance → refundable in full
• consumed balance → non-refundable once API usage has occurred
• promotional / trial credits → non-refundable and non-withdrawable

Token Harbor offers a variety of onboarding rewards, referral incentives, free model access, and limited-time promotional programs. Available rewards and promotions may change over time.

Token Harbor provides free access to selected models for eligible users. Current free access limits:

• Active users — 25 free requests every 5 hours
• Top-up users — 250 free requests every 5 hours

An active user is defined as a user who has completed at least one API request or chat request. Available models, usage limits, and eligibility requirements may change over time.

Every new account receives $5 in trial credit. No credit card is required. The sign-up credit expires 7 days after account registration.

New users can receive a 100% bonus on their first wallet top-up. For example: top up $10 → receive $10 bonus; top up $50 → receive $50 bonus; top up $100 → receive $100 bonus.

The offer is available for 14 days after account registration. Maximum bonus is $100 per account.

Users can earn bonus credits through referrals. When someone signs up using your referral link and becomes an active user, you receive a $2 referral bonus credit. An active user is defined as a user who completes at least one API request or chat request.

Referral rewards are currently uncapped, but still count toward the account's lifetime incentive cap. To prevent abuse, Token Harbor may apply activity verification, anti-fraud checks, and referral validation rules.

All promotional and incentive credits combined are capped at $500 lifetime per account. This includes sign-up credits, top-up match bonuses, referral rewards, and future promotional incentives.

Once the lifetime cap is reached, wallet balances can only increase through direct top-ups.

Some rewards, promotional credits, and limited-time offers may include expiration periods depending on the campaign. Examples may include sign-up credits, top-up match bonuses, seasonal promotions, and cashback campaigns.

Specific expiration details are shown in the dashboard or promotion details when applicable.

For paid models, Token Harbor does not retain your prompts or responses. We believe users should have access to frontier AI models without sacrificing privacy.

Paid models — prompts and responses are not retained by Token Harbor, there is no training on customer data, and they are designed for privacy-sensitive workloads.

Free models — some free models may retain data as part of the provider's free usage program. For example, free DeepSeek access may involve data retention by the underlying provider. Because of this, free models are disabled by default, and users must explicitly enable them before use. You can turn this off anytime: turning it off stops future data sharing and disables free models going forward, while data already shared while it was on is retained. This helps ensure users clearly understand the privacy trade-offs before using free models. Please refer to the model details page for the latest data retention and privacy information.

No. Provider keys are stored securely on the backend and are never exposed publicly.

We believe transparency is critical for AI infrastructure. Token Harbor provides visibility into request logs, model usage, token consumption, request costs, and usage analytics.

For TH Orchestra requests, users can also see which model ultimately handled the request and how usage was billed. Our goal is to help users understand what they are using, how much it costs, and how requests are processed.

Token Harbor works with trusted model and infrastructure providers and continuously monitors service health. For many models, multiple upstream providers are available behind the scenes. If one provider experiences issues, requests can automatically fail over to another provider serving the same model.

This helps improve uptime, reliability, and request success rates while maintaining a consistent model experience for users.

We currently provide support through our Discord community (discord.gg/uBTckEReb5), email support (support@tokenharbor.ai), and documentation.

For technical issues, billing questions, API integration help, feature requests, or general feedback, users are encouraged to join our Discord community or contact the Token Harbor team directly.

We welcome feedback from users and developers. Bug reports, feature requests, and product feedback can be submitted through our Discord community (discord.gg/uBTckEReb5) or email support (support@tokenharbor.ai).

Our Discord community is typically the fastest way to reach the team and discuss new ideas with other users.