auto-fallback: try InferencePort, if 429 use KeylessAI

switch to InferencePort direct API

back to KeylessAI (completions.me 401)

read API key from env var with fallback

switch to completions.me with gpt-5.2

compact history to 20 messages: older ones get summarized into context

log all requests with full details

switch to KeylessAI (no key, no limits, no signup)

add web_type tool for filling forms and typing in textboxes

add web_search, web_click tools for interactive browsing

add web browsing: web_fetch tool + tool call loop

restore original request parsing for shortcuts + history

fix shortcuts empty history, add debug logging

add GET /history, handle messages array in shortcuts

separate POST /reset endpoint for clearing history

shortcuts maintains conversation history, reset with {"reset":true}

rename relay_history to history

restore relay_history as string

remove relay_history from response

relay_history as string

relay_history excludes system prompt; API still sends it

shortcuts: send only messages, no system prompt

shortcuts: send all messages without trimming

add relay_history to response

add request logging on failure (method, url, headers, body)

restore default system prompt for /shortcuts

make /shortcuts accept simpler formats (text/message)

add POST /shortcuts endpoint (alias for /relay)

use lightning model (Lightning-Text v2)

fix: use llama-3.1-8b-instant model (llama3-8b-8192 decommissioned)

sync with sebebeb-relay.hf.space: fix API URL to /v1/chat/completions, add debug endpoint, restore OpenAI payload format