Self-hostedOpen source

Turn target accounts into conversations.

Bring your prospects, run email campaigns and handle replies across email and WhatsApp. Use AI where it helps, review what needs your attention, and keep control of your data.

Self-host itSee how approvals work

Free to run on your own servers · Bring your own Anthropic key · Apache 2.0

Work in the dashboard or connect your agent through MCP, the API or the CLI. The same workspace permissions and sending policies apply.

See what needs your attention. Review outreach before it runs.

Keep agent requests and launch checks close to the work. Signed-in members decide approvals and set the policies and grants that govern automation.

How the gate works

Prepare work in one place

Bring your prospect list, prepare email campaigns and use your connected agent to research and draft. Review the audience, messages and sending limits before launch.

An agent never approves itself

Approvals, suppressions and consent are enforced on the server. No API key, OAuth token or MCP tool can approve a request, its own included.

Rules only a member changes

The autonomy policies that say how far an agent may go are set by a signed-in member. No tool, key or token can change them.

Keep conversations moving

Handle email replies and WhatsApp conversations in the inbox. Agent requests follow workspace policy, with member review where required and explicit grants for permitted automation.

How the work runs. Agents do the legwork. People decide what leaves.

01

Connect and warm your mailboxes

Add Google Workspace, Microsoft 365 or any SMTP and IMAP mailbox. Each one checks its domain records and ramps up slowly before a campaign leans on it.

How warmup works
02

Let your Claude do the legwork

Over MCP, your own Claude reads contacts, deals and replies, researches public pages, and drafts campaigns and replies in your workspace. It drafts; it does not send.

Your own Claude
03

Send at a careful pace

A person launches the campaign. Each mailbox then keeps its own daily cap, a gap between sends and its work hours, and each step is sent exactly once.

How sending works
04

Handle every reply in one inbox

Replies from every mailbox land in one inbox, labelled as they arrive. A reply stops that person in every sequence, and a drafted answer can be waiting for you.

The unified inbox

Yours to run. Your servers, your keys, your data.

SalesOS ships as one codebase you run yourself. There is no hosted version to sign up for, and nothing in it needs one. AI features send the text they work on to the model provider you configure (Anthropic by default), and reply labelling sends reply text to its provider.

Run it on your infrastructure

Run SalesOS from source with the local setup guide. Public SalesOS release images are not published yet; the image installer still targets upstream Warmbly releases unless explicitly configured.

Your own AI key

Set your own Anthropic key and pay Anthropic directly. There is no AI markup and no credit meter.

Your data moves with you

Export a whole workspace to an archive and import it on another instance. Nothing depends on a service of ours.

Pre-release. SalesOS release images are not published yet; to run SalesOS today, build it from source.

Who it is for. Small B2B teams who want agents on the legwork and a person on every send.

Founders

Selling a considered purchase yourself, with Claude doing the research and drafting.

Read more

Small sales teams

Share prospects, campaigns, replies and clear rules about what may send.

Read more

Multi-project operators

Several projects or brands, one workspace each, one Claude across them.

Read more

Not a fit today: large sales organisations that need enterprise ABM advertising, seat-based governance across hundreds of reps, or a bought contact database. What is being built next is on the roadmap.

Run it on a server you control.

Self-host it, let your own Claude do the legwork, and keep the decision about what leaves with people.

Self-host it See how approvals work