No More Limits

Flat monthly coding plans priced by concurrency. No token metering, no usage percentages, no surprise bills.

Serving nowdeepseek-v4.1-flashat api.kourier.sh/v1

Works with the coding agents you already use.

Pay for lanes, not tokens.

A plan is a number of requests that can run at the same time. When one finishes, its lane is free again. Nothing is metered.

Starter

Hands-on coding

$20/month

1 lane

No request-volume or token-usage quotas
Up to 1 concurrent request
DeepSeek V4.1 Flash access; new models added as available
OpenAI-compatible API endpoint
Community support via Discord
Usage dashboard and request logs

Pro

Long sessions, heavy use

$50/month

3 lanes

No request-volume or token-usage quotas
Up to 3 concurrent requests
DeepSeek V4.1 Flash access; new models added as available
OpenAI-compatible API endpoint
Community support via Discord
Usage dashboard and request logs

Comparing options? Full plan details

Light trails from traffic on a curving highway at night, rendered as a green dither

DeepSeek V4.1 Flash, live now.

Frontier reasoning. Strong at code, math, and tool use. Call it by name on any plan.

deepseek-v4.1-flash

Context window: 262,144 tokens; Prompt and output combined
Max output: 16,384 tokens; Per response
Endpoints: OpenAI and Anthropic; Chat Completions, Responses, Messages

Release notes

Qwen 3.6 35B (coming soon)
Upcoming lighter-weight complement for smaller, faster jobs.

Point your agent at it.

One OpenAI-compatible endpoint, plus an Anthropic one. Copy a config, add your key, and keep coding.

Base URL https://api.kourier.sh/v1

shell

export OPENAI_API_KEY="sk-bf-..."
export OPENAI_BASE_URL="https://api.kourier.sh/v1"

Quick start guide

~/.omp/agent/models.yml

providers:
  kourier:
    baseUrl: https://api.kourier.sh/v1
    api: openai-completions
    apiKey: KOURIER_API_KEY
    authHeader: true
    models:
      - id: deepseek-v4.1-flash
        contextWindow: 262144
        maxTokens: 16384

Oh My Pi guide

~/.codex/config.toml

[model_providers.kourier]
name = "Kourier"
base_url = "https://api.kourier.sh/v1"
env_key = "KOURIER_API_KEY"
wire_api = "responses"

[profiles.kourier]
model_provider = "kourier"
model = "deepseek-v4.1-flash"

Codex CLI guide

shell

export ANTHROPIC_BASE_URL="https://api.kourier.sh/anthropic"
export ANTHROPIC_AUTH_TOKEN="sk-bf-..."
export ANTHROPIC_MODEL="deepseek-v4.1-flash"
export ANTHROPIC_SMALL_FAST_MODEL="deepseek-v4.1-flash"

Claude Code guide

Small team. Own hardware. Open weights.

Star trails circling over a dark landscape, rendered as a green dither

Open-Weight Models

Permissively licensed, open-weight models. Our goal is the best quality we can serve, at a price we publish up front.

Our Own Clusters

We run our own hardware and do the heavy lifting: batching, kernel work, scale-out, and support for new model architectures.

No Surprises

A small, bootstrapped team. We welcome outside checks on model quality and third-party review of our setup.

Measured, not promised.

We publish real request timings with sample counts and the method behind them. Measurements are observations, not a speed or uptime guarantee.

See the data

Parallel rail tracks converging toward the horizon, rendered as a green dither

Questions, answered.

DeepSeek V4.1 Flash is available today. We are exploring smaller models to behave as a small fast context gatherer.

Two flat monthly plans with no request-volume or token-usage quotas. Starter is $20/month with 1 simultaneous request; Pro is $50/month with 3. No usage credits, rolling quotas, seat fees, or surprise bills.

We do not use your API prompts or outputs to train or fine-tune models. Prompt and response content is excluded from inference gateway request logs. Requests are processed in memory with temporary caching. We retain operational metadata for usage, billing, security, and service reliability.

We do not currently offer a contractual uptime SLA. We are deploying independent monitoring and historical incident reporting for the public status page.

Yes. If you have an open-weights model you want served, contact us. We deploy it on dedicated GPUs with the same API and observability as our standard lineup.

Yes. The endpoint accepts the OpenAI chat completions schema, so any client built for OpenAI works unchanged. Point the base URL at us, set your key, and the same code runs.

Start with one lane.

Starter is $20/month. Switch to Pro when you want three.