NETFORY Client UI Web4 v1.135.118

The first public implementation of a decentralized network NETFORY WEB4.0

README

[1.135.118 / ai-worker 0.13.0 / ntfry-ai 0.2.0] (native chat dialogs Msg::InferChat)

Main-app release marker rolling up the frontend
changes since the last app version: the AI console Speed Badge
(ai-worker 0.10.1), the pipeline slots badge (0.11.0) and the model
catalog marketplace badges live tok/s + token price (0.12.0).

Added

• Native multi-turn dialogs (Msg::InferChat): the OpenAI messages[]
  array now travels to the gateway as structured Vec<ChatTurn> instead of
  a pre-flattened transcript. The gateway renders the history with the
  model's canonical chat template (new Family::render_chat in
  chat_template.rs: ChatML / Llama 3 / Mistral / Zephyr / GLM-4, including
  system prompts and think=false > empty <think></think> block).
• Free prefix caching for chat: reuse is decided by the token prefix
  after rendering (prefix_best_match in run_batch), so the next turn of
  the same dialog automatically hits the shard snapshot and only prefills the
  tail no extra hashing needed, fast TTFT on follow-ups.
• ntfry-ai 0.2.0 mapping: OpenAI messages[] map to ChatTurn with no
  string concatenation; multi-turn / system > InferChat, a single user
  message > plain Infer (compatible with older workers). If the gateway is
  old and InferChat fails, the proxy automatically falls back to a flat
  transcript Infer. docs/NTFRY-AI.md updated.