Pick a chat on the left or start a new one.
Documents
Loading model
P 0
C 0
CTX –
~$0.000
0.0s • 0 tok/s
No chats yet.
enable_thinking to false for non-reasoning mode on some models.uploads/ and are encoded again only when sent to the model.
uploads/ and is encoded again only when sent to the model.
web_search, web_read and image_generate. Default: 20 rounds. 0 = unlimited.
n_ctx with output room reserved. Default: 100,000. 0 disables automatic compaction.
user message.192.168.*/10.*/localhost, etc.
Off: private networks stay blocked.
/search and /news.
X-API-KEY.
X-Subscription-Token.
Authorization: Bearer ….
{base}/search and {base}/news)
or a full endpoint URL (e.g. index.php) that is called directly.
/search/index.php) does not allow automatically deriving a news route.
X-API-KEY.
{base}/search with format=json.
Important: in your SearXNG instance, JSON output must be enabled (search → formats → json), otherwise no JSON will be returned.
X-API-KEY.
ragd. LAN addresses are supported; no localhost assumption is built into LLC.
ragd for each RAG-enabled prompt. Default: 8.
ragd. Default: 180 seconds.
/v1/images/*. OpenRouter uses /chat/completions with image output. stable-diffusion.cpp uses its OpenAI-compatible /images/generations endpoint. ComfyUI uses your own server (workflow JSON).
https://api.openai.com/v1.
Authorization: Bearer ...
/models. OpenAI returns basic model metadata only, so image-capable IDs are selected by naming pattern.https://openrouter.ai/api/v1. Models are loaded from /models?output_modalities=image.Authorization: Bearer ...1234. You may enter either http://host:1234 or http://host:1234/v1; requests go to /v1/images/generations. A browser GET on /v1 may show “error”; use /v1/models as a basic check.sd_cpp_extra_args. This is the escape hatch for flags the OpenAI API shape does not know.image_edit tool becomes available to the AI only when a valid edit profile is configured here.Create OpenAI-compatible function tools backed by HTTP endpoints, without editing JSON for the common cases.
Connect Streamable-HTTP MCP servers and expose their tools to the existing agentic tool loop.
gpt-4o-mini-tts (stable).
Manage OpenAI-API compatible endpoints (e.g. openrouter.ai, api.openai.com). Stored locally.
Admins can create users, grant/revoke admin rights, and change email/password.
LocalLightChat - the high performance chat frontend.
Project URL: https://www.locallightai.com/llc/
© 2026 SRWare. All rights reserved.
{"arguments": ...} as JSON.{{parameter_name}} placeholders. If the rendered template is not valid JSON, it is sent inside a fallback JSON envelope.mcp_local__fs_read.