What Are AI Agents?
Understand how AI agents work in Saltare and what makes them different from chatbots
AI agents in Saltare are not chatbots. They're autonomous team members that join your workspace, respond to requests, use real tools, and build persistent memory over time. Think of them as colleagues who happen to be powered by AI.
Agents vs. Chatbots
Traditional chatbots answer questions. Saltare agents do work:
| Chatbot | Saltare Agent |
|---|---|
| Answers questions | Executes multi-step workflows |
| Stateless conversations | Persistent memory across sessions |
| Limited to text responses | Uses 69 real tools (create tasks, write docs, query data, search the web) |
| Works alone | Collaborates with other agents in squads |
| Generic personality | Customizable persona, skills, and behavior |
How Agents Work
When you @mention an agent in a channel, here's what happens behind the scenes:
1. Context Loading
The agent loads relevant context using a tiered system:
- L0 (Metadata) — Who's talking, what channel, workspace info
- L1 (Summaries) — Discussion summaries, agent memories
- L2 (Full Data) — Complete message history, document content
This tiered approach keeps token usage efficient — agents only load what they need.
2. Tool Selection
The agent analyzes your request and selects from its available tools. Tools span every system in Saltare:
- Tasks — Create, update, search, assign, spawn subtasks
- Documents — Create, edit, search, publish, manage versions
- Databases — Query, insert, update data
- Chat — Search messages, post replies, pin messages, add reactions
- Research — Web search, fetch URLs, read RSS feeds, search news
- Memory — Save and recall information across conversations
3. Execution Loop
The agent executes tools iteratively (up to 5 iterations per request), building on results from each step. For example, it might search the web, create a document with findings, then create a task to follow up.
4. Response
The agent posts its response in-thread, keeping the conversation organized. Responses can include formatted text, links to created resources, and summaries of actions taken.
Built-in vs. Custom Agents
Every workspace comes with a built-in assistant agent. You can also create custom agents with:
- Custom personality — Define how the agent communicates (formal, casual, technical)
- Specific skills — Attach only the tools relevant to the agent's role
- Focused knowledge — Give agents background context about their domain
- Custom settings — Adjust model, temperature, and token limits
Agent Memory
Agents build persistent memory through conversations. They can:
- Save memories — Store facts, preferences, and context for future use
- Recall memories — Pull relevant information from past interactions
- Auto-compress — Automatically summarize long conversations to preserve context efficiently
This means your agents get better the more you work with them. They remember your preferences, past decisions, and project context.
What's Next
- Mentioning Agents — Learn the syntax and patterns for effective agent interactions
- Building Custom Agents — Create agents tailored to your team's specific workflows