bcat — admin

LLM Provider

Model

Active Model

Status Checking...

Chat Theme

Choose a color theme for the chat interface. Each theme is inspired by a different cat breed.

Model Management

Installed Models

Checking...

Pull New Model

API Keys

OpenAI Key

OpenAI Base URL

Anthropic Key

Google AI Key

Ollama Settings

Ollama URL

System Prompt Override

Custom system prompt (leave blank for default)

Advanced Features

Cat Personality Level

🤖 🐱 5

Balanced — 1-2 cat touches per response

Fast Model

Multi-model routing: simple questions use the fast model, complex queries use the main model. Leave blank to disable.

Vision Model

Image understanding: uploaded images are described by this model. Must be an Ollama vision model.

Memory & Learning

Persistent memories extracted from your conversations

Loading...

📥

Drop files here or click to upload

Images, documents, code — the agent will read and remember the contents

📄 file.txt

Uploading...

✓ Ingested

Learning Trend

Loading…

Pending Requests

Loading...

Users

Loading users...

Service Control

Restart Bcat Services

Restarts bcat.service and ollama.service only. Does not affect nginx, docker, or other shared services.

API Clients

Provision Bearer tokens that authorize external apps to call /api/v1/* endpoints. Each token has its own scopes and rate limit. Tokens are shown once at creation — only the SHA-256 hash is stored, so a leaked DB cannot leak the secret. Revocation is instant.

Name

Description

Scopes

* — grant all scopes — wildcard. Token can call every public endpoint, including any added in the future. Still rate-limited and audited; does not grant infer (admin-only). Use this when issuing a token to an app you fully trust and don't want to re-issue every release. ask — smart-routed Q&A. Caller sends a question; bcat picks the best specialist (code/math/biology/etc.) and returns the answer. Recommended for most apps. embed — vector embeddings (default model nomic-embed-text). Use for semantic search, RAG, or similarity scoring. models — read-only list of models available on this Ollama instance. Lets clients self-discover capabilities; no inference cost. vision — image understanding. POST /api/v1/vision (multipart) sends an image to the configured vision model and returns a description. Useful for OCR, diagram analysis, alt-text generation. extract — document text extraction. POST /api/v1/extract (multipart) returns plain text from PDFs, source code, JSON/XML, markdown, etc. No model call — fast and cheap.

The infer scope (raw model passthrough, bypasses routing) is admin-only and not selectable here.

Rate / minute Use 0 or unlimited for no per-token limit.

Monthly cap Total requests/month. Blank = unlimited.

Provisioned Clients

API Activity

Self-Study

When idle, bcat picks an under-used learned pattern, asks itself a probe question to test or refine it, answers, then runs that answer through the qwen3:8b judge. Only refinements scoring at or above the threshold are kept (as self_studied memories). Every probe is logged below — accepted, rejected, or otherwise — so you can audit exactly what bcat is teaching itself.

Health Notifications

Recipients (comma-separated)

One or more addresses. Leave throttle = off to disable all notifications.

Throttle

Recovery/test emails are never throttled. Throttle applies to down/restart events.

SMTP Transport

SMTP Host

SMTP Port

SMTP User

SMTP Password

From Address

Server Logs

Connecting... 0 entries