Repository navigation
Fixed
-
A slow local model gets ten times the wait, and response.patience changes it — #2854 ·
enginechatdocs4 things that are no longer true
- A model on your own machine or network (Ollama, LM Studio, llama.cpp, a box on the LAN or Tailscale) was held to the hosted figures: 2 minutes for response headers, 10 seconds of silence after headers, 45 seconds quiet mid-reply, a 5-minute reply wall, 150 seconds for a non-streamed reply, and a 2-minute compaction summary. Under the default
response.patience: autoa local endpoint now gets ten times each of those, and the turn's give-up stretches with them; hosted endpoints keep the old figures. - A slow request to a local endpoint was sent again to the same server after 10 seconds, so one GPU did the work twice. Under
autoa local endpoint is never sent a duplicate unless the gateway lists more than one provider for the model. - Nothing could make codeaf wait longer for a quiet model;
response.attemptsonly scaled the give-up.response.patience(envCODEAF_RESPONSE_PATIENCE, thepatiencerow on the Models tab) takesautoor a number from 1 to 100 that applies to every endpoint the same way, and1is exactly the behaviour before this change. A bad value is refused naming the row or the variable. - A local server that sent its headers at once and its first token after a long prompt read was cut at the action deadline (100 seconds under
auto) and asked again, every time, so a first token later than that never arrived. Underautoa request to a local endpoint that lists no second provider is never cut to ask the same machine again; it waits for the first-token limit, 15 minutes on a conversation turn.
A local model can spend minutes reading a long prompt before its first token,
and it is the only machine there is to ask. "Local" is decided from the address
alone: loopback, private, carrier-grade and link-local addresses,.local,
.lanand.internalnames,host.docker.internal, a single-word host, or the
Ollama connection wherever it runs. A saved row reaches an open conversation on
its next model request. - A model on your own machine or network (Ollama, LM Studio, llama.cpp, a box on the LAN or Tailscale) was held to the hosted figures: 2 minutes for response headers, 10 seconds of silence after headers, 45 seconds quiet mid-reply, a 5-minute reply wall, 150 seconds for a non-streamed reply, and a 2-minute compaction summary. Under the default
Install this stable build:
curl -fsSL https://raw.githubusercontent.com/Agent-Field/CodeAF/1a682dfa838ff56042b0499b48d13b46214e7157/scripts/install.sh | bash -s -- --stable
curl -fsSL https://raw.githubusercontent.com/Agent-Field/CodeAF/1a682dfa838ff56042b0499b48d13b46214e7157/scripts/install.sh | VERSION=v0.8.3 bash