Projects

A project is the top-level container in Bonsai. It represents one complete conversational AI experience — everything from the stages and agents to the conversations and users lives inside a project.

Creating a Project

Go to Administration > Projects and click Create Project.

Basic Settings

• Name — A human-readable name (e.g., "Customer Support Bot", "Sales Assistant").
• Description — Optional notes about the project's purpose.

Voice Settings

Setting	Description
Accept Voice	Whether the project accepts voice input from users
Generate Voice	Whether the project generates spoken audio output (TTS)

If voice is enabled, you'll also need to configure:

ASR Configuration (Speech-to-Text)

Controls how user speech is transcribed:

• ASR Provider — Select a registered speech recognition provider (set up in Providers).
• Unintelligible Placeholder — Text used when speech can't be transcribed (default: [unintelligible]).
• Voice Activity Detection — Automatically detect when the user starts and stops speaking.

Server-side VAD

Experimental Feature

Server-side VAD is under active development. Behaviour may change in future releases.

Enable Server-side VAD switches from client-driven voice input (where the client explicitly signals start/end of speech) to continuous audio streaming with automatic speech boundary detection on the server.

When enabled, the following settings become available:

Setting	Description	Default
Aggressiveness Mode	How aggressively non-speech audio is filtered out (0 = least aggressive, 3 = most aggressive)	2
Frame Duration	Duration of each VAD processing frame. Must be 10, 20, or 30 ms	20 ms
Silence Pre-roll Padding	Silence prepended before detected speech as a pre-roll buffer (0–1000 ms)	300 ms
Auto-End Silence Duration	How long silence after speech must persist before end-of-utterance is triggered (100–5000 ms)	800 ms

Server-side VAD is useful when clients cannot reliably detect speech boundaries themselves (e.g. embedded devices or browser environments without access to audio APIs).

Storage Configuration

Experimental Feature

Storage is under active development. Behaviour may change in future releases.

Controls where conversation artifacts (audio recordings, transcripts, images) are saved:

• Storage Provider — Select a registered storage provider (S3, Azure Blob, Google Cloud Storage, etc.).

If no storage is configured, only inline data is preserved.

Timezone

Set a project-wide default timezone (IANA format, e.g., Europe/Warsaw, America/New_York). This is used in prompt templates via the time context and can be overridden per-conversation or per-user.

Language Code

Set an ISO language code for this project (e.g., en-US, pl-PL, de-DE). This is exposed in the conversation context as project.languageCode (the raw code) and project.language (a human-readable name such as "American English"). Use it in prompts or scripts to instruct the AI to respond in a specific language.

Conversation Timeout

Set a maximum inactivity period (in seconds) for active conversations. If a conversation receives no user input for longer than this value, it is automatically aborted. Accepted range is 60–3600 seconds. Leave the field empty to disable the timeout entirely.

Constants, Memory, and Moderation

Project-level constants, user-profile memory schema, and moderation settings are configured in the Design section. This keeps all conversation-design configuration in one place.

Child Resources

A project contains:

• Stages — Conversation phases
• Agents — AI personalities and voices
• Classifiers — Intent detection
• Context Transformers — Data extraction
• Tools — AI-powered functions
• Knowledge Base — FAQ categories and items
• Global Actions — Shared action definitions
• Guardrails — Project-wide safety rules
• Global Memory — User profile schema and project constants
• Moderation — Content moderation configuration
• API Keys — Client authentication credentials
• Conversations and Users (visible in the Monitor section)

Tips

• One bot = one project — Keep each conversational experience in its own project.
• Configure storage early — If you want to keep audio recordings and transcripts, set up storage before going live.
• Set a timezone — This ensures the AI gives accurate date and time information in conversations.