TXT CLAW Routing (Hosted Router Lane)
TXT CLAW supports two LLM lanes:
- Hosted router (default): TXT CLAW pays the model bill and routes requests.
- BYOK (optional): you bring a provider key (see ).
https://www.txtclaw.com/byok.md
This doc describes the hosted router behavior and the
llm
Hosted Router Defaults
- :
llm.mode
(default)hosted
- :
llm.tier
(default)fast
- : optional explicit override
llm.model
Auto-Router (Primary + Fallback)
When configured, TXT CLAW can try a primary model and fall back to a smarter model on upstream failures.
How it works (conceptually):
- Attempt 1: primary model (fast)
- Attempt 2+: fallback model (smart), if configured
Create An Agent With Hosted Router
bash
export TXTCLAW_API_BASE_URL="https://txtclaw-sms-e2e.lopez731.workers.dev"
export TXTCLAW_API_KEY="vck_REPLACE_ME"
curl -sS "$TXTCLAW_API_BASE_URL/v1/agents" \
-H "Authorization: Bearer $TXTCLAW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"system_prompt": "You are a helpful assistant. Keep replies concise.",
"llm": { "mode": "hosted", "tier": "fast" },
"sms": { "mode": "none" }
}'Force A Specific Model (Advanced)
bash
curl -sS "$TXTCLAW_API_BASE_URL/v1/agents" \
-H "Authorization: Bearer $TXTCLAW_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"system_prompt": "You are a helpful assistant.",
"llm": { "mode": "hosted", "model": "openai/amazon/nova-lite" },
"sms": { "mode": "none" }
}'Notes
- The exact hosted models are an implementation detail; use unless you need a specific override.
tier
- If you’re wrapping TXT CLAW, treat as an advanced escape hatch, not a default.
llm.model