Built-in workforce

Your team starts with
13 workers.

Every Salsa instance ships with pretrained agents that handle real work — research, outreach, compliance, recruiting, and more. They run inside the same execution loop as your human teammates: QA-gated, sandbox-isolated, audit-trailed.

Always improving
Team-assignable
Sandbox-isolated
Custom agents

Not chatbots. Workers.

Agents that execute, not suggest

Pretrained on real workflows

Each agent ships with a playbook — a structured set of instructions honed on actual business processes. Not generic prompts. Real operational knowledge.

Constantly evolving

Agent personas are updated with every Salsa release. New capabilities, better playbooks, refined heuristics. Your workforce gets smarter without any action from you.

Same rules as humans

Agents go through the same QA gates, permission checks, and audit logging as your human teammates. No shortcuts. No shadow execution.

Pretrained workers

Ready to work on day one

Assign any of these agents to your team. They operate on schedules, respond to events, and execute inside the Salsa loop. Activate the ones you need. Suspend the rest.

PM

Project Manager

Weekdays at 10:00

Tracks milestones, manages blockers, and keeps teams aligned

An always-on coordinator that monitors task completion, identifies dependency risks, gathers status updates from every channel, and delivers progress reports. Thinks in milestones, not messages.

Playbook Track milestones · Monitor dependencies · Flag blockers · Send progress reports · Maintain documentation
PW

Pipeline Whisperer

Weekdays at 11:50

Keeps your CRM healthy and nudges reps at the right time

Surfaces stale deals, drafts follow-up messages, syncs meeting notes into opportunity records, and forecasts close probability. Flags any deal with no activity in 5+ days.

Playbook Review CRM daily · Flag stale deals · Draft follow-ups · Sync call notes · Update forecasts
Event trigger New meeting notes (call transcript added)
CH

Customer Health Monitor

Weekdays at 10:05

Surfaces churn risk and expansion signals before anyone else

Combines product usage analytics, support ticket velocity, and sentiment feeds to score account health. Alerts CSMs on usage drops and compiles weekly executive summaries.

Playbook Score account health · Alert on usage drops · Track support tickets · Weekly executive summaries
Event trigger Critical support ticket opened
CI

Competitor Intelligence Analyst

Weekdays at 13:20

Monitors rivals, product launches, and pricing moves

Aggregates press releases, product changelogs, community chatter, and job postings into concise competitive briefs. Tracks pricing changes, executive hires, and sentiment shifts across the market.

Playbook Monitor pricing · Track product updates · Flag executive hires · Sentiment analysis · Competitive briefs
Event trigger Competitor press release or RSS alert
TS

Talent Scout

Tuesdays at 14:30

Finds and qualifies candidates across LinkedIn, GitHub, and job boards

A 24/7 recruiting partner. Define skills, seniority, and location. Talent Scout discovers matching profiles, captures background and project history, and delivers structured candidate dossiers.

Playbook Search LinkedIn/GitHub/boards · Qualify candidates · Capture experience · Deliver dossiers
Event trigger New job requisition approved
OC

Onboarding Concierge

Weekdays at 16:00

Welcomes new hires and keeps the checklist moving

Guides every new teammate through the entire onboarding journey: welcome messages, equipment provisioning, orientation scheduling, paperwork collection, and stakeholder nudges for overdue items.

Playbook Welcome notes · Confirm equipment · Schedule orientation · Collect paperwork · Nudge stakeholders
Event trigger HRIS marks new employee hired
VP

Vendor Price Analyst

Monday & Thursday at 15:05

Tracks supplier quotes and finds negotiation leverage

Compares your contracted vendor pricing against public catalogs, competing quotes, and historical trends. Surfaces renegotiation opportunities and recommends tactics when price changes exceed 3%.

Playbook Collect quotes · Monitor catalogs · Highlight drift >3% · Recommend negotiation tactics
Event trigger New invoice uploaded by finance
CS

Compliance Sentinel

Mondays at 12:10

Audits policies and alerts owners when controls drift

Keeps SOC 2, ISO 27001, and internal policy tasks on track. Diffs policy repos, checks evidence freshness, reminds control owners of upcoming deadlines, and summarizes audit readiness.

Playbook Review control evidence · Flag missing docs · Track audit windows · Summarize readiness
Event trigger 30 days before external audit window
IC

Incident Comms Scribe

Every hour, every day

Captures status updates and keeps stakeholders aligned during incidents

Records incident timeline updates, drafts stakeholder emails, pulls PagerDuty and Slack threads into a single narrative, and ensures post-mortem materials have a clean, auditable history.

Playbook Capture updates · Draft stakeholder emails · Build timeline · Prepare post-mortem materials
Event trigger PagerDuty incident opens
RE

Real Estate Research Analyst

Daily at 09:00

Finds properties, pulls comps, and tracks market trends

Monitors real estate listings, researches comparable properties, analyzes market data, and compiles reports. Delivers email updates with new opportunities and trend analysis.

Playbook Monitor listings · Research comps · Analyze market data · Compile property reports
EA

Enrichment Agent

On demand

Adds context and supporting data to any record or task

Enriches incoming records with relevant metadata, highlights missing fields, and assigns confidence levels. Works across CRM entries, support tickets, lead records — any structured data.

Playbook Enrich records · Source metadata · Highlight gaps · Score confidence
OA

Outreach Agent

On demand

Drafts personalized outreach and coordinates sends

Writes outreach messages tailored to each recipient, proposes optimal send windows, and maintains auditable logs of every touchpoint. Every draft passes QA review before send.

Playbook Draft messages · Personalize · Propose send windows · Keep auditable logs
SC

Salsa Crawler

On demand

Crawl, scrape, map, and extract structured web data

A full web intelligence worker. Four modes: map (discover URLs), scrape (fetch and clean to markdown), crawl (recursive traverse), and extract (schema-validated JSON). Enforces robots.txt and rate limits. Preserves provenance.

Playbook Map URLs · Scrape pages · Crawl recursively · Extract structured data · Enforce robots.txt
Event trigger Team dispatches crawl/scrape/extract task

Execution model

Every agent runs in the loop

Agents don't bypass safety. They follow the same five-stage execution loop as every human action in Salsa.

1

Trigger

A schedule fires, an event arrives, or a human assigns work. The agent receives a structured intent.

2

Plan

The agent decomposes the task into steps using its playbook. Each step specifies tools, inputs, and expected outputs.

3

QA Gate

An independent QA agent reviews the plan for safety, completeness, and risk. No plan executes without passing.

4

Sandbox Execute

Steps run inside an isolated VM sandbox. The host system is never exposed. Failed steps can't damage your environment.

5

Audit & Notify

Every action is recorded in an immutable event store. Results are delivered with a full trail of what happened and why.

Orchestration modes

One agent. Or all of them.

Salsa supports three orchestration modes depending on the complexity of the work.

Single

One agent, one task

Route a task to a specific agent. Best for focused, well-defined work like drafting outreach or monitoring compliance.

Example "Outreach Agent, draft a follow-up for the Acme deal"
Plan → Execute

Decompose, then dispatch

A plan agent breaks the task into steps, QA reviews the plan, then each step routes to the right specialist agent automatically.

Example "Research competitor pricing, enrich our CRM records, draft an executive brief"
Swarm

Parallel execution, synthesized results

Multiple agents work simultaneously on different facets of the same problem. Results are merged into a unified output.

Example "Prepare the board deck: financials + market intel + customer health + recruiting pipeline"

Marketplace agents

Domain-specific agent teams

Beyond the core workers, Salsa ships specialized agent teams for entire business functions. Each team is a coordinated group of agents with shared context.

Growth Pipeline

A complete marketing team in a box

16 coordinated skills that take you from competitor research to live ad campaigns.

Research
Ads Analyst Meta Ads Extractor Landing Page Analyst Ad Creative Analysis
Strategy
Head of Marketing Brand Analyst Campaign Planner
Create
Creative Director Ad Designer Scriptwriter Page Designer
Publish
Performance Marketer Meta Ads Publisher

Custom agents

Build agents for your exact workflow

When the built-in workers don't fit, build your own. Custom agents inherit the full Salsa execution model — QA gates, sandbox isolation, audit trails, event triggers, and team assignment.

Define a playbook

Write a structured playbook that describes the agent's purpose, operating procedures, success criteria, and failure modes. The playbook is the agent's operational DNA.

Set triggers and cadence

Configure when the agent runs: scheduled (cron-style), event-driven (webhook, status change), or on-demand. Mix triggers to cover every scenario.

Assign to teams

Bind the agent to specific teams with scoped permissions. Each team gets its own instance with isolated state. Suspend or activate per team without affecting others.

Long-horizon work

Custom agents excel at multi-day, multi-week projects. They maintain context across sessions, remember previous runs, and build on accumulated knowledge — not just single-shot tasks.

Under the hood

System agents that power the loop

Behind every worker is a foundation of system-level agents that handle reasoning, planning, quality assurance, and tool execution. You never configure these — they just work.

Reasoning

LLM Agents

Claude, GPT-4, Ollama, llama.cpp, and Apple MLX for on-device inference. Salsa's tiered routing sends 90% of work to fast/cheap models and only escalates complex reasoning to frontier models.

5 providers Tiered routing Local-first
Planning

Plan Agent

Decomposes complex tasks into directed acyclic graphs of steps. Each step specifies which tool to use, what inputs to provide, and what success looks like.

DAG decomposition Tool routing System trust
Quality

QA Agent

The mandatory quality gate. Reviews every plan for safety, completeness, and risk before any execution happens. Cannot be bypassed. Cannot be overridden.

Mandatory Pre-execution Non-negotiable
Execution

Terminal & Code Agents

Execute shell commands, generate code, and run build tools — all inside the VM sandbox. Elevated trust level with full audit logging of every command.

Sandboxed Elevated trust Full audit

Continuous improvement

Agents that get better every release

1

Playbook refinement

Every release includes updated playbooks based on real-world usage patterns and edge cases discovered across the Salsa user base.

2

New capabilities

Agents gain new tool access, integration endpoints, and orchestration patterns. New pretrained workers are added based on the most common custom agent patterns.

3

Model upgrades

When frontier models improve, Salsa's tiered routing automatically benefits. Smarter reasoning at the same cost. Better quality without changing your configuration.

4

Community patterns

Successful custom agent patterns from the community get promoted to pretrained workers. The best ideas from power users become built-in for everyone.

Start with a full team.

Every Salsa instance ships with 13 pretrained workers, system agents, and the ability to build custom agents for any workflow. Request beta access to put them to work.