Kiln AI

PUBLIC BETA · CLOUDFLARE EDGE

Five capabilities.
One clear endpoint.

No sign-up or API key. Try Kimi, GLM, GPT-OSS, and Gemma with SSE streaming and two OpenAI-compatible APIs.

Start a chat View integration
Auth
Anonymous beta
Transport
SSE only
Runtime
Cloudflare Edge
CFWORKERS
AI
KIMIK2.6KIMIK2.7 CODEGLM5.2GPTOSS 120BGEMMA4 26BFIVE MODELS · ONE EDGE

02 / MODEL SHELF

Pick for the task.

AAVAILABLE

MOONSHOT AI

Kimi K2.6

Long context, vision, reasoning, and general coding.

  • Multi-turn chat
  • Vision and reasoning
  • Streaming output
kimi-k2.6Workers AI
BAVAILABLE

Z.AI

GLM 5.2

Chinese, complex reasoning, and agentic coding workflows.

  • 262K context
  • Reasoning and functions
  • Strong Chinese support
glm-5.2Workers AI
CNEW

MOONSHOT AI

Kimi K2.7 Code

Agentic coding with vision, structured output, and multi-turn tools.

  • 262K context
  • Function and tool calls
  • Code-first workloads
kimi-k2.7-codeWorkers AI
DNEW

OPENAI

GPT-OSS 120B

An open-weight model for reasoning, agents, and developer workloads.

  • 128K context
  • Reasoning
  • Function calling
gpt-oss-120bWorkers AI

03 / CONNECT

One SSE stream.
Two open APIs.

Chat Completions supports common apps; Responses API supports Codex and agent tooling. Streaming responses use SSE only.

POST/api/v1/chat/completions
POST/api/v1/responses
GET/api/models
REAL CMD TEST · HTTP 200
Click terminal to copyCMD before the API response
01 Request entered

04 / ENDPOINT GUIDE

How to use every endpoint.

GET /api/models

Lists aliases, context windows, output ceilings, and compression thresholds.

curl https://api-ai.js.gripe/api/models

POST /api/v1/chat/completions

Works with common OpenAI chat clients. Set stream: true for SSE.

curl https://api-ai.js.gripe/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k2.6","messages":[{"role":"user","content":"Hello"}],"stream":true}'

POST /api/v1/responses

Works with Codex and Responses clients using a string or input-item array.

curl https://api-ai.js.gripe/api/v1/responses \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-oss-120b","input":"Hello","stream":true}'

GET /api/tools/web-search

Reads the tool and skill manifests. Responses uses {"type":"web_search"}; Chat can set web_search: "auto".

curl https://api-ai.js.gripe/api/tools/web-search
curl https://api-ai.js.gripe/api/skills/web-search

App setup: use https://api-ai.js.gripe/api/v1 as the Base URL; no API key is required. Use stream: false for JSON or stream: true for SSE. In auto, the selected model decides whether to search from its model-specific knowledge-cutoff system prompt; third-party requests that declare no tool default to off. Send a stable x-session-affinity for each conversation to maximize native Workers AI prefix-cache hits; x-kiln-prompt-cache reports affinity status. The API is stateless and never silently compresses caller context; use thresholds from /api/models if your client manages memory. See /openapi.json for the complete schema.

05 / HOW IT RUNS

Static front. Edge back.

  1. 01
    Pagekiln / Pages

    Static pages ship quickly while chat state stays in the tab.

  2. 02
    API Worker

    SSE, Chat, Responses, and model-decided search.js.gripe web search; WAF handles traffic policy.

  3. 03
    Workers AI

    Models run at the edge; credentials never reach the browser.

ENEW

GOOGLE

Gemma 4 26B A4B

An efficient MoE open model for vision, reasoning, structured output, and tools.

  • 256K context
  • Vision and function calling
  • Native Workers AI prefix caching