Building Custom Agents

While every workspace comes with a built-in assistant, the real power of Saltare is in creating custom agents designed for specific roles on your team. A custom agent can be a researcher, writer, analyst, project manager, or any role your team needs.

Creating an Agent

Navigate to your workspace settings to create a new agent. You'll configure:

Name and Identity

• Name — How the agent is addressed (e.g., "Researcher", "Editor", "Analyst")
• Description — A brief summary of what the agent does, visible to team members
• Avatar — A visual identity for the agent in chat

System Prompt

The system prompt defines the agent's personality, expertise, and behavior. This is the most important configuration — it shapes how the agent thinks, responds, and approaches tasks.

A good system prompt includes:
- Role definition — "You are a senior research analyst specializing in competitive intelligence"
- Behavioral guidelines — "Always cite your sources. Present findings in structured tables when comparing options."
- Domain knowledge — "Our company sells B2B SaaS tools for HR teams. Our main competitors are..."
- Constraints — "Never make up data. If you can't find reliable information, say so explicitly."

Skills

Skills determine what tools the agent can use. Three types:

• Tool skills — Invoked through chat (e.g., web search, document creation)
• Trigger skills — Activated by events (e.g., new task created, webhook received)
• Automation skills — Run on schedules (e.g., daily report generation)

By default, agents have access to all workspace tools. You can restrict this to only the tools relevant to the agent's role.

Agent Design Patterns

The Researcher

Role: Deep research and analysis
Skills: web_search, fetch_url, search_news, create_document, search_messages
System prompt focus: Thorough sourcing, structured output, citation standards

The Project Manager

Role: Task management and team coordination
Skills: create_task, update_task, search_tasks, list_channels, post_message
System prompt focus: Prioritization frameworks, status tracking, deadline awareness

The Writer

Role: Content creation and editing
Skills: create_document, update_document, search_documents, web_search
System prompt focus: Writing style, brand voice, editorial standards

The Analyst

Role: Data analysis and reporting
Skills: query_data_table, natural_language_query, create_document, web_search
System prompt focus: Statistical rigor, visualization preferences, insight extraction

Multi-Agent Collaboration

Agents can work together on complex workflows. When one agent uses the collaborate tool, it can invoke another agent in a hidden thread, passing context and receiving results.

Example workflow — competitive intelligence:

1. Researcher discovers all online mentions of a competitor
2. Analyst evaluates each mention for sentiment and impact
3. Writer synthesizes findings into an executive briefing document

This pattern — discover, fan-out, synthesize — generalizes to any multi-step research workflow.

Advanced Settings

For fine-tuning agent behavior:

• Model — Choose between different Claude models (Sonnet for speed, Opus for depth)
• Temperature — Lower (0.3) for factual tasks, higher (0.8) for creative work
• Max tokens — Control response length
• Context budget — How much workspace context the agent loads per request

Best Practices

• Start simple — Begin with a focused role and expand capabilities as you learn what works
• Test in a dedicated channel — Create a test channel to iterate on your agent's system prompt before deploying to the team
• Review agent memories — Periodically check what your agents have memorized to ensure accuracy
• Iterate on the system prompt — The system prompt is a living document. Refine it based on the agent's actual performance