A sales agent can research a lead, draft a follow-up and write the result to the CRM with nobody watching. A message that reaches a buyer is the place to put a person. Kortix is the open-source AI Operating System that runs the agent, and its approval gate is a policy in the repo you own.
One stops a send, one decides who may merge, one governs a single call.
The agent researches and drafts freely, but nothing goes out on its own: every follow-up stops at a human approval gate as a draft for a person to review and send.No message reaches a lead without a person reviewing the draft and sending it.Only the drafts it produces are written back out.The starter agent ships with kortix_permissions: all. Drop project.gitops.merge from that list and its change requests wait for a person.
kortix_version: 2default_agent: kortixagents: kortix: file: agents/kortix.md connectors: all kortix_permissions: allThe first rule that matches wins. A send that matches the pattern runs; the same action to any other address waits for a person. A changed argument needs a new approval.
policy: default_mode: risk policies: - match: gmail.send_email action: always_run conditions: - arg: to match: "*@example.com" - match: gmail.send_email action: require_approvalThe change request is where a person reads the work before it lands.
The agent commits, pushes the session branch, and opens a change request against the project's default branch.A person merges itA person with the project.gitops.merge permission reviews the change request and merges it into the default branch.Source: https://kortix.com/docs
Each one sits in a place a reviewer opens before the first send.
A kortix_permissions list on the agent decides what it may do, and it sits in kortix.yaml where a reviewer reads it before the first run.
Session work arrives as a change request, so the review is a diff of the files the agent touched. What lives in that repo covers the layout.
Credentials are encrypted in the secrets manager and injected at runtime, scoped to the agents you grant them to. Connector credentials are brokered server-side and never enter the machine.
What each tool does on its own, and where the control over a send sits.
| Product | Open source | What runs unsupervised | Where the approval control lives | What a person sees afterwards | What you change without the vendor |
|---|---|---|---|---|---|
| Kortix | Yes | Research, drafting, CRM notes, a slot; you set the send gate | A policy and a permission list in your git repo | Every action is recorded; reading the audit log depends on the plan | The persona, skills, permission list and tool rules in the repo |
| AiSDR | No | Builds and runs campaigns, sends, classifies replies, books meetings | In its platform: pause outbound at any time | A unified inbox holding every email and LinkedIn message | Messaging, ICP and sequences inside its product |
| Artisan | No | Finds leads, writes outreach, handles replies, books meetings | In its platform: approve before sending, lock tone and CTAs | Audit logs, held on its platform | Ava’s coaching and guardrails inside its product |
Rival cells come from each vendor’s own pricing page, checked October 2026. Kortix cells come from its published lead follow-up flow and its docs. The Kortix vs AiSDR page takes the managed-SDR trade-off further.
Short answers on what an agent may do alone, and on running the gate yourself.
Yes. Agents research the lead, write the pre-call brief, update the CRM, draft the follow-up and propose a slot, and the lead follow-up flow Kortix publishes holds the send for a person. The comparison hub sets that flow beside the managed SDRs.
Yes. Kortix runs on Kortix Cloud, in your VPC, or on your own on-prem network, and self-host is free, so the gate and the repo stay on your infrastructure.
Start with new-lead follow-up and one tool rule on the send step.