Four AI agents for your Salesforce org, with a human in every write.
The Salesforce Agent Platform connects to your org and your repo, then gives your teams agents for understanding, building, selling, and supporting on top of Salesforce.
Four agents, one platform
Each agent is built on the same shared plumbing: organizations, permissions, connections, the LLM layer, and background jobs.
Org Intelligence
Builds a living picture of your Salesforce org and metadata repo: objects, automation, Apex and LWC, permissions, integrations, and drift between org and repo.
- Health score and tech-debt findings
- Org vs. repo drift detection
- Versioned snapshots you can diff
Dev Agent
Designs and builds business automation and UI, choosing between Flow and Apex, scaffolding Apex, LWC, and Flow, and running static analysis before anything is reviewed.
- Design, then approve, then generate
- PMD, ESLint, and Code Analyzer checks
- Opens pull requests. Never deploys directly
Sales Agent
Pipeline insight for reps and managers: forecast rollups, deal risk flags, coaching, and grounded follow-up drafts over your playbooks and deal data.
- Forecast and win-rate figures computed, not guessed
- Deal risk rules with reasons
- Drafts only. Nothing is sent or written to Salesforce
Service Agent
Case triage, sentiment, knowledge suggestions, and response drafting, with escalation driven by rules you configure.
- Structured triage and escalation rules
- Replies grounded in your knowledge base
- Read-only access to cases. Never writes to Salesforce
Build your own agents and workflows
The four built-in agents are a starting point. Your team can design agents and multi-step flows that fit your processes, using the same tools, permissions, and approvals.
Custom agents
Admins configure chat agents for a specific job, choose the tools they can use, and publish them to the people who need them.
- Role-based access per agent
- Tools limited to what you allow
- Usable in the console and in Slack
Agentic workflows
Design multi-step flows that combine tool calls, agent steps, and conditions, with triggers and an approval step where a person must sign off.
- Visual workflow designer
- Approvals with named approvers and timeouts
- Runs tracked step by step
RAG-grounded agents
Load your documents, runbooks, and enablement content into vector collections, then let agents and workflow steps search them before they answer.
- Collections from files, the API, or a workflow
- Vector search steps inside flows
- Answers tied to the source material
Connects to the tools you already use
Agents work where your engineering, delivery, and sales teams already work. Each connection uses scoped, encrypted credentials.
Git
GitHub repositories
Connect a repository to analyze your Salesforce metadata, track drift against the org, and have the Dev Agent open pull requests on new branches.
Jira
Jira Cloud
Start an agent request from a ticket, post progress back, and let the platform update ticket status as work moves through review.
Slack
Channels and direct messages
Get notifications, ask the agents questions, and approve workflow steps from Slack without switching tools.
Salesforce
OAuth and JWT
Read orgs, metadata, and pipeline data through each user's own connection, so Salesforce permissions still decide what each person sees.
Built for security and control
Agents are only useful if your admins and security team can trust them. These are the guarantees the platform is designed around.
Permissions come from Salesforce
Sales and Service reads run through the user's own Salesforce connection, so sharing rules, role hierarchy, and field-level security still apply.
Tenant isolation by design
Every organization's data is row-isolated, and the role is re-read on every request. Removing a member takes effect on their next request.
Humans approve every write
Write actions wait for a person to approve their exact inputs. Changes ship as reviewed pull requests, never as direct deployments.
Model-agnostic, Claude by default
A provider-agnostic LLM layer runs Claude by default, with tool calls validated against schemas and every model call metered per organization.
How it fits together
Your people reach the platform through the console, Slack, or Claude Code. One core runs every agent, workflow, and search, and reaches your data and tools through scoped connections.
The platform underneath
The agents share one foundation, so new capabilities arrive as a new agent or tool rather than a new product. Deployment runs on Google Cloud, with the web console on Firebase Hosting.
- Multi-tenant organizations with role-based access
- Salesforce (OAuth) and GitHub connections, encrypted at rest
- Background jobs with progress tracking
- Retrieval (RAG) over your knowledge with pgvector
- Agent chat console and analytics
- Audit log and rate limits on every tenant
See the agents on your own org
Tell us which Salesforce processes you want to speed up, and we'll show you where the agents fit.