KeepRouter

KeepRouter is an independently operated API gateway that provides one account for models from multiple providers. Supported chat models accept OpenAI Chat Completions requests at /v1/chat/completions and Anthropic Messages requests at /v1/messages. Its production catalog lists 101 models, public customer prices, and supported endpoints across text, image, video, and speech output, with pricing units that vary by modality. The model named free costs $0.0000 per 1M input tokens and $0.0000 per 1M output tokens; it supports chat completions, streaming where supported, and tool calling where supported. Paid usage draws on prepaid USD credit at published per-model rates, with no recurring subscription. Keys can be limited to model allowlists, and the raw key is shown only once when created. Request logs include operational details such as route, status, token counts, cost, cache state, and latency, but not prompt or completion bodies. KeepRouter says customer data is not used to train models. Response caching is disabled by default. The service runs on Cloudflare's global edge network and documents setup paths for OpenCode and Continue using its OpenAI-compatible base URL.

KeepRouter offers a unified API gateway with published model pricing and no recurring subscription for paid usage. Review the per-model rates and save each API key securely when it is created.

KeepRouter suits developers who want to access models from multiple providers through one API gateway. It may also suit users of OpenCode or Continue who can configure those tools with its OpenAI-compatible base URL.

KeepRouter is a cloud API gateway that puts models from multiple providers behind one account. It is best for developers who want a shared integration, visible per-model rates and prepaid usage rather than a subscription. Its breadth and security controls are useful, but buyers who need durable free capacity or prompt-content inspection should choose another tool.

Overview

KeepRouter’s production catalog contains 101 models from multiple makers, with customer pricing and supported endpoints shown for each model. That makes it practical to compare text, image, video and speech options before changing an application. It is an access layer rather than a full business-operations analytics system: consumer and latency analytics are present, while business activity metrics are not.

Requests run on Cloudflare’s global edge network. KeepRouter says customer data is not used to train models, and business customers can contact support about a DPA. Paddle acts as Merchant of Record for top-ups, invoices and tax, which centralizes payment administration.

Key features

Familiar API routes

Supported chat models accept OpenAI Chat Completions requests at /v1/chat/completions. They can also accept Anthropic Messages requests at /v1/messages, including use with Claude Code and the Anthropic SDK. This can reduce migration work for teams already built around either interface, although endpoint support is model-specific and must be checked before selection.

Multimodal catalog and free model

The catalog spans text, image, video and speech output, with billing units that differ by modality. A model named free charges $0.0000 per 1M input tokens and $0.0000 per 1M output tokens and supports chat completions, streaming where supported, and tool calling where supported. It is useful for evaluation and prototypes, but its best-effort status makes it a poor basis for a critical production dependency.

Usage visibility without content capture

Request logs record the timestamp, model, route, HTTP status, token counts, cost, cache state and latency. Prompt and completion bodies are not stored, so teams can investigate spend and performance without creating a transcript archive; teams that need to review message content will find this boundary limiting.

Keys, transport and caching

Keys can be restricted to model allowlists, helping separate applications or environments by permitted models. The raw sk-kr- key appears only once at creation, while stored keys are SHA-256 hashes, so losing the original value means creating a replacement. API authentication uses HTTPS with TLS. Response caching is off by default; if enabled, identical non-streaming requests may be cached transiently in Cloudflare KV, a setting that deserves care for responses that should not be reused.

Developer integrations

The Free model documentation provides configuration paths for OpenCode and Continue using KeepRouter’s OpenAI-compatible base URL. Support is available at [email protected], including vulnerability reports and DPA discussions.

Pricing

Free

The Free plan costs free, billed at $0 per input and output token. There is no monthly fee. It is explicitly best-effort and aimed at testing, prototyping and evaluation, so it fits experimentation rather than workloads that require dependable capacity.

Pay-as-you-go

The paid plan has custom pricing for the plan itself, with usage deducted from prepaid USD credit at each model’s published rate and no recurring subscription. Text is billed per token; image models are billed per image or token; video and speech output are billed by duration. This structure suits variable usage and lets teams pay for selected models, but it requires topping up credit and monitoring different units across modalities. There is no free trial.

Platforms

KeepRouter is delivered through its cloud service and exposes API and web access. Its Cloudflare edge deployment is relevant to applications that need a globally distributed gateway, while the web catalog and logs support model and usage administration.

Who it's for

Choose KeepRouter if you maintain applications that may switch among providers, want OpenAI- or Anthropic-compatible routes, or need published rates and scoped keys without a recurring platform fee. It also fits teams that value cost and latency records while keeping prompt and completion bodies out of logs.

Look elsewhere if your service depends on guaranteed free capacity, a conventional monthly entitlement, business-activity reporting, or content-level observability. The best-effort free model and prepaid accounting demand operational discipline.

Pros and cons

• Pro: One account reaches 101 models from multiple makers, including text, image, video and speech options.
• Pro: OpenAI Chat Completions and Anthropic Messages routes can preserve existing client integrations.
• Pro: Published per-model rates and modality-specific billing make usage costs inspectable before deployment.
• Pro: Model allowlists, one-time key display, SHA-256 storage and TLS provide concrete access controls.
• Pro: Logs expose cost, tokens, latency and cache state without retaining prompt or completion bodies.
• Con: The Free plan is best-effort, so it is unsuitable as a production capacity promise.
• Con: Prepaid credit and differing units for text, images, video and speech complicate budgeting.
• Con: Business activity metrics are absent, and content-free logs cannot support prompt-level investigations.

Alternatives

For analytics centered on API traffic, browse API Analytics Software. Apirelio is a web freemium option with a 14-day trial and paid Growth and Scale plans for event volumes, projects, retention and members. Clavio suits readers seeking dataset analysis and charting rather than a model gateway. Helicone is worth considering when an open-source, API or self-hosted analytics deployment is preferred. Langfuse offers API, self-hosted and web deployment with usage quotas, retention and user limits. Moesif is an API and self-hosted analytics alternative with a trial and event-based plans. Revenium targets transaction observation with a free Observe tier and enterprise scale. AI Nazar provides an open-source, MIT-licensed SDK and dashboard. API Analytics is a free API, self-hosted and web option with up to 1 million stored requests per user.

Verdict

KeepRouter is a strong choice for developers who want one gateway to many models, compatible chat APIs, scoped credentials and transparent usage billing. Choose it for flexible, usage-based model access and operational cost visibility; choose another service when reliable free capacity, recurring-plan simplicity, business metrics or prompt-content logging matters more.