FlashAPI

Developer preview · updated 27 September 2026

One chat API for multiple model choices

FlashAPI is designing an OpenAI-compatible text chat endpoint for teams serving Southeast Asia. This page describes the current pilot contract so you can judge whether it fits your client.

Early access statusThe gateway is in internal testing. No external model route is available to customers, no customer API key has been issued, and the example below cannot be called yet.

Proposed endpoints

During a pilot, we will issue a base URL and a client key. Send the key in Authorization: Bearer <key>. One model ID selects a configured provider while your application keeps the same request path.

GET/v1/models

Lists only models actually configured for your pilot. An unconfigured model will not be advertised.

POST/v1/chat/completions

The planned pilot accepts text messages and returns JSON. Streaming with stream: true depends on testing a supported upstream route.

The interface covers a subset of OpenAI chat completions. Provider-specific features will be added only after they can be tested and documented.

Request example for a future pilot

FLASHAPI_BASE_URL and FLASHAPI_API_KEY are values a pilot customer would receive. The model ID here is provisional; the live model list, once available, is authoritative.

curl "$FLASHAPI_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer $FLASHAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-flash",
    "messages": [{"role": "user", "content": "Summarize this in Bahasa Indonesia."}],
    "max_tokens": 256,
    "stream": false
  }'

Use stream: false for a single JSON response. If streaming is enabled for your pilot route, read the text/event-stream response incrementally. Client behavior and optional fields must be checked against the actual upstream chosen for that pilot.

Initial text-chat support

Planned pilot request support
FeaturePilot contract
MessagesText-only system, user, and assistant roles
OutputJSON; streaming SSE only after route-specific validation
Controlsmax_tokens or max_completion_tokens, temperature, top_p, and stop
ReasoningOnly reasoning_effort: "none" in this low-latency pilot
AvailabilityUnknown or unconfigured model IDs receive a clear error. No provider is available today.

Images, audio, tool calls, JSON schema output, automatic fallback, and a multi-provider SLA are outside this initial contract. Ask us about the feature your workload needs before building against it.

What happens next

  1. Share your target country, languages, model tasks, token volume and response-time goal.
  2. We validate upstream access, commercial terms, data handling and the exact model version.
  3. We issue pilot credentials and publish measured latency and pricing for that route before inviting production traffic.

Use the provider-direct migration checklist before a pilot ↗

See the reproducible latency test method and runnable script ↗

Tell us what your API client needs

We are prioritizing the first integrations using real workloads in Indonesia, Vietnam, Thailand and nearby markets.

Request early access