Customer-controlled keys
Your DeepSeek key stays in the router process on your machine or server. Agents receive only the TEKIZ.AI endpoint and optional client key.
DeepSeek BYOK
Put one or more DeepSeek API keys on your TEKIZ.AI host, point Codex
or another OpenAI-compatible client at port 8790, and
keep your provider key off the client. Add another authorized key or
model if you want fallback.
Your DeepSeek key stays in the router process on your machine or server. Agents receive only the TEKIZ.AI endpoint and optional client key.
A 401, 402, 403, or 429 cools the affected credential fingerprint. The router can try another eligible route if you configured one.
Use deepseek/deepseek-v4-flash directly or let Balanced
and Frontier profiles choose among authorized routes.
DEEPSEEK_API_KEY on the machine running TEKIZ.AI
Router, then start the router on port 8790.
http://127.0.0.1:8790/v1. The
provider key does not need to be copied into Codex or another
agent.
Run these commands from your configured router installation. Leave the router running, then send the test request from a second terminal.
export DEEPSEEK_API_KEY=your_deepseek_key python3 router.py --port 8790
curl http://127.0.0.1:8790/v1/chat/completions \
-H 'Authorization: Bearer local-router' \
-H 'Content-Type: application/json' \
-d '{
"model": "deepseek/deepseek-v4-flash",
"messages": [{"role": "user", "content": "Reply with DEEPSEEK_OK"}]
}'
Teams can configure an ordered pool for separate accounts, quota domains, or key rotation. When one key fails, the router can try another eligible key in the pool.
export TEKIZAI_ROUTER_DEEPSEEK_API_KEYS='["team-a-key","team-b-key"]' export TEKIZAI_ROUTER_CREDENTIAL_STRATEGY=failover
Supported strategies are failover,
round-robin, lru, and random.
export TEKIZAI_API_KEY=local mkdir -p ~/.codex cat >> ~/.codex/config.toml <<'TOML' [model_providers.tekizai-local] name = "TEKIZ.AI" base_url = "http://127.0.0.1:8790/v1" env_key = "TEKIZAI_API_KEY" wire_api = "responses" TOML # ~/.codex/tekizai-deepseek.config.toml cat > ~/.codex/tekizai-deepseek.config.toml <<'TOML' model_provider = "tekizai-local" model = "deepseek/deepseek-v4-flash" TOML codex --profile tekizai-deepseek
DeepSeek currently documents V4 Flash as its Codex-compatible model.
Plain codex keeps using ChatGPT. The TEKIZ.AI route is
selected only with codex --profile tekizai-deepseek.
A generated sk_tekiz_* key uses hosted capacity enabled
for your TEKIZ.AI account. A
DEEPSEEK_API_KEY environment variable belongs on your
self-hosted router; setting it on a client does not upload it to the
hosted service. See the
integration matrix for BYOK, BYOS, local,
and managed-access boundaries.
If your company approves only DeepSeek, keep that boundary. TEKIZ.AI can reduce token burn by choosing eligible models. Your DeepSeek account handles provider billing.