AI Chat for Projects — Staff Guide

AI Chat for Projects — Staff Guide

The AI Chat module adds an AI assistant to every project. Staff can hold saved, shareable conversations with an AI model in the context of a specific project, attach documents or images for the AI to read, and export any conversation to PDF. Each project can be given its own instructions and its own AI provider.

1. What it does

• Project-aware chat. Open a project and use the AI Chat tab to chat with the assistant. Conversations are saved as named sessions and shared with everyone on the project.
• Per-project rules. Each project can define its own instructions (a system prompt), provider, model, temperature, and response length — similar to ChatGPT Project settings or Claude Project instructions.
• Multiple providers. OpenAI (GPT), Claude, Gemini, Perplexity, and Meta/Llama are all supported. An administrator supplies the API keys once; each project then picks which provider it uses.
• Attachments. Attach a PDF, Word file, text file, or image to a message and the assistant reads it in its reply.
• PDF export. Export a conversation to download it or attach it to the project's Files tab as part of the record.

2. One-time administrator setup

An administrator configures providers once, from Setup → AI Chat (or directly at /admin/openai_chat/settings).

1. Enter the API key for each provider you want available.
2. Optionally set a default model for each provider. Leave blank to use that provider's built-in default. Model fields are free-text because provider model names change often.
3. Set the global defaults: default provider, default instructions, temperature, and max response tokens. Any project that hasn't set its own values inherits these.

OpenAI key: if a key already exists in the CRM's core AI settings, leave the module's OpenAI field blank and it inherits that key. Fill it in only to override.

Meta / Llama: this provider needs a Base URL pointing at an OpenAI-compatible host (for example Groq, Together, or Fireworks), because there is no single first-party Meta endpoint.

3. Setting a project's AI rules

1. Open the project and click the AI Chat tab.
2. Click Project AI Settings.
3. Choose the provider and, optionally, a specific model.
4. Write the instructions — how the assistant should behave for this project (its role, tone, focus, and any constraints).
5. Optionally adjust temperature (focused vs. creative) and max response tokens (reply length).
6. Save.

Anything left blank inherits the system-wide defaults, so you only set what matters for that project.

4. Using the chat

1. On the AI Chat tab, create a New Session and give it a name (for example "Requirements" or "Client Objections").
2. Type a message and press Enter. The assistant replies in the thread.
3. Use the paperclip to attach a document or image for the assistant to read.
4. Sessions are shared, so a colleague can open and continue any session on that project.
5. Use Export PDF to download the conversation or attach it to the project's Files tab.

5. Good to know

• Instructions apply to new sessions. Changing a project's instructions affects sessions created afterward. Sessions already in progress keep the instructions they started with — the same way ChatGPT and Claude projects behave.
• Image support varies by provider. OpenAI, Claude, and Gemini read attached images. Text-only endpoints (some Llama hosts) receive a text note in place of the image.
• If a provider errors, confirm its API key is set on the settings page (and, for Meta/Llama, that the Base URL is set).

6. Who does what

• Administrator: enters provider API keys and global defaults, once, under Setup → AI Chat.
• Project members: set a project's instructions and provider on the AI Chat tab, then create sessions and chat.