Connect Codex, OpenClaw, SDKs, and coding tools to the hosted API or a router you run yourself. The hosted API uses a TEKIZ.AI key. A local router keeps provider credentials on the machine you control. Find the setup steps for each tool below.
Use https://api.tekiz.ai/v1 with an active
sk_tekiz_* key. Your account shows usage limits and
request history; the service can try another eligible route after
a retryable failure when one is available to your account.
Use http://127.0.0.1:8790/v1 for tools running beside
your router. Provider credentials and supported subscriptions stay
on the router host you control. Ask about installation access
if you need a local router.
Use http://<tailnet-host>:8790/v1 from trusted
devices on your private Tailnet.
Keep your default Codex setup unchanged and opt into TEKIZ.AI with
a named profile. The guide writes the provider to
~/.codex/config.toml and the profile to
~/.codex/tekizai.config.toml.
Aider uses ~/.aider.conf.yml; Continue uses
~/.continue/config.yaml; and OpenClaw uses
~/.openclaw/openclaw.json. Each guide includes a
copy button and a key placeholder. Never paste a real key into a
repository file.
Cursor uses its Settings interface. Claude Code and Kimi use shell environment variables. DeepSeek Harness reads its native settings file. Follow the linked guide for the tool you use.
Use tekizai/auto first. Select
tekizai/frontier or tekizai/fusion when
that profile fits the task. Set the client key (e.g.
TEKIZAI_API_KEY or KIMI_API_KEY) in the
shell, OS secret store, or service environment; do not commit a key.
Keep the default provider unchanged and use a named Responses API profile.
# ~/.codex/config.toml [model_providers.tekizai] name = "TEKIZ.AI" base_url = "https://api.tekiz.ai/v1/" env_key = "TEKIZAI_API_KEY" wire_api = "responses" # ~/.codex/tekizai.config.toml model_provider = "tekizai" model = "tekizai/auto" codex --profile tekizaiOpen the full Codex guide
Merge this provider under models.providers, then permit its models for the agent.
"tekizai": {
"baseUrl": "https://api.tekiz.ai/v1",
"apiKey": "${TEKIZAI_API_KEY}",
"api": "openai-completions",
"models": [
{ "id": "tekizai/auto", "name": "TEKIZ.AI Auto" },
{ "id": "tekizai/frontier", "name": "TEKIZ.AI Frontier" },
{ "id": "tekizai/fusion", "name": "TEKIZ.AI Fusion" }
]
}
Open the full OpenClaw guide
Use the Anthropic-compatible endpoint. It has no /v1 suffix.
ANTHROPIC_BASE_URL=https://api.tekiz.ai
ANTHROPIC_API_KEY=${TEKIZAI_API_KEY}
Open the full Claude Code guide
Set the API key in Hermes’s service environment and configure its OpenAI-compatible provider.
model: provider: tekizai default: tekizai/auto base_url: https://api.tekiz.ai/v1
Pi resolves this environment reference at request time.
{
"providers": {
"tekizai": {
"baseUrl": "https://api.tekiz.ai/v1",
"api": "openai-completions",
"apiKey": "${TEKIZAI_API_KEY}",
"models": [{ "id": "tekizai/auto", "name": "TEKIZ.AI Auto" }]
}
}
}
Merge this OpenAI-compatible provider adapter into the OpenCode configuration.
{
"provider": {
"tekizai": {
"npm": "@ai-sdk/openai-compatible",
"name": "TEKIZ.AI",
"options": {
"baseURL": "https://api.tekiz.ai/v1",
"apiKey": "{env:TEKIZAI_API_KEY}"
},
"models": { "tekizai/auto": { "name": "TEKIZ.AI Auto" } }
}
}
}
Add this custom OpenAI-compatible provider in ZCode’s provider settings.
{
"provider": {
"tekizai": {
"name": "TEKIZ.AI",
"kind": "openai-compatible",
"options": {
"baseURL": "https://api.tekiz.ai/v1",
"apiKey": "${TEKIZAI_API_KEY}",
"apiKeyRequired": true
},
"source": "custom",
"models": { "tekizai/auto": { "name": "TEKIZ.AI Auto" } }
}
}
}
Kimi Code uses a dedicated coding endpoint and API key. It speaks
the Anthropic Messages API at https://api.kimi.com/coding/.
ANTHROPIC_BASE_URL=https://api.kimi.com/coding/
ANTHROPIC_API_KEY=${KIMI_API_KEY}
DeepSeek Harness reads its provider row from
~/.dsh/settings.yaml. Point the
llm-deepseek adapter at the hosted endpoint and set
DEEPSEEK_API_KEY to a non-empty TEKIZ.AI key in the
service environment.
# service environment
DEEPSEEK_API_KEY=${TEKIZAI_API_KEY}
# ~/.dsh/settings.yaml
llm-deepseek:
baseURL: https://api.tekiz.ai/v1
api: openai-completions
models:
- id: tekizai/auto
name: TEKIZ.AI Auto
contextWindow: 131072
maxTokens: 8192
agent-default-model:
provider: deepseek-official
model: tekizai/auto
Choose the hosted or local endpoint first. Keep account and provider credentials out of support messages.
Set
OPENAI_BASE_URL=https://api.tekiz.ai/v1 or
local port 8790, then call
tekizai/frontier through chat completions or
Responses API-compatible routes.
Use wire_api = "responses",
model = "tekizai/frontier", and the hosted,
local, or Tailnet base URL where you run TEKIZ.AI.
Point the tool at your hosted account or local router endpoint. Check the available models before selecting a profile.
Run TEKIZ.AI beside Ollama or another model service. It finds available models where supported, while your tools connect only to TEKIZ.AI.
Keep one or more DeepSeek API keys on the router host, isolate failures per credential, and use V4 Flash directly or through a route profile. Open the DeepSeek guide.
Use the Anthropic-compatible endpoint when your client expects it. On a local router, provider credentials stay on the host you control.
Use the native configuration file for each agent and keep
TEKIZAI_API_KEY in its environment or secret store.
Start with tekizai/auto, then select
tekizai/frontier or tekizai/fusion
for the task and available capacity.
Save authorized provider API keys on the computer running your router. Supported provider subscriptions can use their approved sign-in method. Local Ollama usually needs no key.
You can add more than one key for the same provider. If one fails, the router can pause that key and try another authorized key. Subscription sign-ins work only through integrations approved by that provider.
When asking for help, share your account email, plan, time of the problem, HTTP status, request ID, error type, and only the first few characters of a key. Never send prompts, API keys, passwords, cookies, model responses, or customer data.