New: Voice calls with AI agents / Learn more
Features / AI Agents

AI team members
that actually work.

Not chatbots. Not copilots. Autonomous agents with real tools, persistent memory, and the ability to collaborate with each other. Summon them with an @mention.

Free plan includes 2 agents and 50 credits/month

#competitive-intel
you
Researcher Searching 6 sources... Found 18 mentions across news, blogs, and social. Creating research document.
Analyst Scoring 18 mentions. Spawning subtask per source. ETA: 3 minutes. Will post synthesis when complete.
The Old Way
Chat with an AI, copy the output, paste it somewhere else
Re-explain context every single conversation
One AI assistant for your entire organization
AI can talk but can't actually do anything
With Saltare Agents
Agents live in your workspace — they create tasks, write docs, and update databases directly
Persistent memory across conversations — they learn and improve over time
Specialized agents for different roles — Researcher, Analyst, Writer, PM
69 tools across tasks, docs, research, databases, and integrations
Capabilities

Not a chatbot. A coworker.

Every agent comes equipped with real tools, persistent memory, and the autonomy to chain complex workflows.

01 // Tool Arsenal

69 tools across 11 categories

Agents don't just generate text. They execute actions — creating tasks, searching the web, writing documents, querying databases, and pulling data from connected services.

Tasks
create_task update_task search_tasks spawn_subtasks
Research
web_search fetch_url search_news lookup_person
Documents
create_document update_document search_documents
Chat
post_message search_messages list_channels
Memory
save_memory recall_memories
Integrations
github_* slack_* google_* discord_*
02 // Persistent Memory

Context that compounds

Agents remember past conversations, learn your preferences, and build up institutional knowledge. A tiered loading system keeps token usage efficient — agents load only what they need.

L0 // Metadata Always loaded

Agent identity, workspace info, channel context, member list

L1 // Summaries Loaded on demand

Agent memories, discussion summaries, recent activity context

L2 // Full Data Budget-gated

Complete message history, document contents, full database records

03 // Multi-Agent Squads

Agents that collaborate

Agents invoke each other via the collaborate tool in hidden threads. A Researcher finds data, an Analyst scores it, a Writer drafts the report — all triggered by a single @mention.

R
Researcher collaborate(Analyst, "score these 18 mentions")
A
Analyst collaborate(Writer, "draft executive summary")
W
Writer create_document("Competitive Intel — Q1 2026")
04 // Skills & Customization

Tailored to your workflow

Attach modular Skills to agents to extend their abilities. Three skill types cover every automation pattern:

T
Tools

Chat-invoked abilities. "Summarize this thread" or "Search our database for leads in NYC."

E
Triggers

Event-driven reactions. When a GitHub issue is opened, an agent triages and assigns it automatically.

A
Automations

Scheduled via cron. "Every Monday at 9am, compile last week's metrics and post a summary."

Execution Pipeline

From @mention to finished work

What happens under the hood when you mention an agent.

01

Mention Detection

Your message is parsed for @mentions. Each agent mention creates a record and enqueues a response job.

02

Context Assembly

The ContextBuilder loads tiered context — agent identity, memories, discussion history — within an 8,000-token budget.

03

Agent Loop

The agent calls the AI model, selects tools, executes them, and iterates up to 5 times until the task is complete.

04

Response & Follow-up

Results are posted as a message in-thread. If the agent invoked collaborators, their work chains automatically.

Use Case

Competitive intelligence pipeline

One @mention triggers a multi-agent research workflow. Discover, fan out, synthesize.

01

Discovery

You mention @Researcher with "Find all online appearances for Jane Doe." The agent sweeps search engines, social platforms, press, podcasts, and public records.

web_search"Jane Doe" site:linkedin.com
search_news"Jane Doe" mentions last 90 days
fetch_urlcrunchbase.com/person/jane-doe
create_document"Jane Doe — Appearances (18 found)"
02

Deep Dive

The agent spawns a subtask for each appearance. Multiple agents work in parallel — visiting each source, extracting key quotes, scoring sentiment, and identifying related entities.

TechCrunch Interview Complete
LinkedIn Profile Complete
Forbes 30 Under 30 Processing
Podcast Ep. #247 Queued
03

Synthesize

A final agent rolls up all subtask outputs into a comprehensive dossier — timeline, key themes, network graph, and recommended actions. Posted to your channel and saved as a document.

S
Synthesizer Just now
Research complete. Created Jane Doe — Intelligence Dossier.
18 appearances analyzed. Sentiment: 72% positive. 3 recommended follow-up actions attached as tasks.

Stop describing work.
Start delegating it.

Your first AI agent is ready to go in under a minute. No training required.

Deploy Your First Agent

Not ready yet? Get updates when we launch new features.