Harbourlight Flash

AI model tokens and engineering services

AI model tokens, and the engineers who put them to work.

Harbourlight Flash sells usage-based access to AI models through one API, including the models AWS serves through Amazon Bedrock. Our forward-deployed engineers build with your team, and our consultants tune AI workloads for cost and quality.

Operated by Harbourlight Flash Limited, a company incorporated in Hong Kong.

Client

Your applications

Products, internal tools and data pipelines

Gateway

Harbourlight gateway

  • One key for every model
  • Per-token metering
  • Spend limits

Model sources

    Fig. 1Requests from your applications reach the gateway through one API. The gateway meters each request and forwards it to the model you chose, on the provider's own API or on AWS through Amazon Bedrock.

    Example route

    01Tokens

    One key and one bill for the models you use.

    Buy tokens for models from several providers through a single account. You integrate once, pay for what you use, and see what every request cost.

    • Unified API

      One endpoint and one key for models from several providers, in the OpenAI-compatible format most SDKs already support. Changing models is a one-line change.

    • Models served on AWS

      We also supply the models AWS offers through Amazon Bedrock, including the Amazon Nova family. Call them with the same key and the same per-token billing.

    • Spend limits

      Give each project or environment its own API key and set a budget on it. A key stops working when its budget is used up.

    • Usage records

      Each request is recorded with its model, token counts and cost, so you can attribute spend to a team, a product or a feature.

    02Services

    Engineers and advice for teams building on these models.

    Token supply covers access. These two services cover the work around it: building the system, then making it cheaper and better once it runs.

    Forward-deployed engineering

    Our engineers join your team and ship in your codebase.

    • Scope one use case with your product and data owners, then build it end to end
    • Connect models to your systems: retrieval over your documents, tool calls and internal APIs
    • Write evaluation sets and tests, so each change is checked before release
    • Hand over working code, documentation and a runbook to your engineers

    AI optimization consulting

    Lower cost and better output from the workloads you already run.

    • Break down token spend by model, feature and prompt to show where the money goes
    • Match each task to a suitable model, and send simple requests to cheaper ones
    • Cut tokens per request with prompt caching, shorter context and batch jobs
    • Measure quality, latency and cost before and after each change

    How an engagement runs

    1. Intro call

      You describe the goal, the systems involved and your timeline.

    2. Written scope

      We send a statement of work with deliverables, duration and price.

    3. Delivery

      We work in your environment and report progress at agreed checkpoints.

    4. Handover

      You receive the code, documentation and results, and your team takes over.

    03Pricing

    Tokens billed by usage. Services quoted by scope.

    Token charges follow usage: the tokens your requests send to a model and the tokens it returns. Engineering and consulting are priced per engagement.

    1. 1

      Per-token rates

      Every model has an input rate and an output rate, each quoted per 1 million tokens. A request costs its input tokens times the input rate, plus its output tokens times the output rate.

    2. 2

      Prepaid credit or monthly invoice

      Accounts add prepaid credit and draw it down as they use the API. Approved business accounts can instead receive a monthly invoice for the previous month's usage.

    3. 3

      Rate card with your quote

      Rates follow provider pricing and can change. We send the current rate card for the models you plan to use, and your usage records show the rate applied to each request.

      Ask for the rate card
    4. 4

      Services by statement of work

      Forward-deployed engineering and consulting are quoted per engagement, by scope and duration, in a written statement of work.

    04Security and data handling

    Clear rules for your keys, your data and our access.

    The same rules apply whether you buy tokens from us or bring our engineers into your systems.

    • Encryption in transit

      API traffic is served over TLS, and requests to model providers travel over encrypted connections.

    • Request logging you choose

      Billing needs only token counts, the model and the time of each request. Tell us whether to keep request content for debugging or metadata only.

    • Key management

      Issue a key per project or environment, set a spend limit on each, and revoke or replace any key at any time.

    • Provider restrictions

      Limit an account to the providers and models you approve, for example AWS-served models only, so traffic follows your own data policies.

    • Work inside your environment

      On engineering engagements we work in your repositories and cloud accounts with the access you grant. That access ends at handover.

    • Confidentiality

      We sign a confidentiality agreement before we see your code, data or prompts.

    Each request is processed by the model provider or cloud service you route it to, under that provider's terms for API data.

    05Company

    The company behind the service.

    Harbourlight Flash Limited is a private company limited by shares, incorporated in Hong Kong on 9 October 2026. We sell access to AI models, billed by usage, and provide forward-deployed engineering and AI optimization consulting to businesses.

    For quotes, contracts, invoices and account questions, write to xgyw88888@163.com.

    Model, provider and cloud service names on this site belong to their owners. Harbourlight Flash is an independent company and is not affiliated with or endorsed by any model provider or by Amazon Web Services.

    Company record
    Legal name
    Harbourlight Flash Limited
    Company type
    Private company limited by shares
    Jurisdiction
    Hong Kong SAR
    Business registration number
    81351921
    Incorporated
    9 October 2026
    Registered address
    Room B21, 6/F, Po Kuan Building, 50 Hung To Road, Kwun Tong, Hong Kong

    06Contact

    Talk to sales.

    Tell us which models and volumes you need, or describe the system you want built or tuned. We reply by email.

    Registered address
    Room B21, 6/F, Po Kuan Building, 50 Hung To Road, Kwun Tong, Hong Kong

    Sales enquiry

    We use these details only to reply to you. See the Privacy Policy.