JanitorAI

Allows non-technical users to define AI character personalities, conversation styles, and behavioral constraints through a web-based form interface without writing code. The system likely parses natural language character descriptions and system prompts into internal configuration objects that seed the underlying LLM's behavior, enabling rapid prototyping of custom chatbots with minimal technical friction.

Implements automated content filtering on bot-generated responses to prevent unsafe, inappropriate, or policy-violating outputs before they reach users. The system likely uses a combination of keyword filtering, pattern matching, and potentially classifier models to detect and block or sanitize responses containing violence, sexual content, hate speech, or other flagged categories, with configurable sensitivity levels per bot.

Maintains conversation history across multiple exchanges, allowing bots to reference prior messages and build context for coherent long-form dialogue. The system manages a rolling context window (likely 4K-8K tokens) that includes recent conversation history, character definition, and system prompts, feeding this context to the LLM for each new response generation to maintain conversational continuity.

Provides serverless hosting for created chatbots with automatic scaling, uptime management, and no infrastructure setup required from users. Bots are deployed as web-accessible endpoints (likely REST APIs or WebSocket connections) that handle concurrent user conversations, with the platform managing load balancing, database persistence, and availability without exposing infrastructure details to creators.

Provides a structured interface for defining character traits, speech patterns, knowledge domains, and behavioral rules that are compiled into system prompts injected into the LLM context. Users select or write character attributes (e.g., 'sarcastic', 'knowledgeable about history', 'avoids political topics') which are translated into natural language instructions that guide the model's response generation, enabling consistent personality without fine-tuning.

Enables creators to publish bots to a platform directory with shareable links, allowing other users to discover, interact with, and potentially fork or remix existing characters. The system likely maintains a searchable/browsable catalog of public bots with metadata (creator, description, interaction count) and provides URL-based sharing for direct access without requiring directory discovery.

Exposes bot functionality via REST API or webhooks, allowing external applications to trigger bot conversations, retrieve responses, or receive notifications of user interactions. The system likely provides authentication (API keys), rate limiting, and structured request/response formats (JSON) for programmatic bot access, enabling integration with Discord bots, Slack workspaces, or custom applications.

Tracks and displays metrics on bot usage, user engagement, and response quality, providing creators with insights into how their bots are performing. The system likely logs conversation metadata (message count, session duration, user retention) and may provide dashboards showing popularity trends, user feedback, or response satisfaction scores to help creators iterate on bot design.