Kortix vs Vellum: a personal assistant you shape, or an open-source system your team owns

Vellum is a personal AI assistant one person names and shapes, and Kortix is the open-source system a team configures together in a repo it owns. Both run on your own hardware or in a managed cloud, and both let you bring your own model. The question is who the assistant belongs to: a single person who teaches it, or a whole company that configures it.

What Vellum gets right

Vellum is a personal assistant before it is anything else, and these are the parts it does well.

• An assistant that becomes yours. Vellum takes a personality you give it. During onboarding it watches how you communicate and writes its own personality files, and its homepage puts it plainly: you shape every aspect of it, from the way it talks to the way it works (Vellum).
• Memory built around one person. Vellum keeps 8 types of memory about you, the people around you and your work, so it carries context across conversations instead of starting cold each morning (Vellum Assistant).
• A safe default on risky actions. Sending mail to a new contact or spending credits needs your approval until you mark the action trusted, and its credentials sit in a separate process the model never reaches (Vellum).
• One identity on every surface. The same assistant and the same memory reach macOS, iOS, Android, the web app, voice, email, Telegram, Slack and the terminal, so a task started in one place continues in another (Vellum Assistant).
• Models and hosting you choose. Switch between OpenAI, Anthropic, Gemini and open-weight models through Ollama, run it locally, or let Vellum Cloud run it for you (Vellum).

Where one assistant stops

Vellum is sized for a single person, and that shape shows as soon as more than one person has to steer it.

• The unit is a person, not a team. The identity, the memory and the plan all orbit one user. Its paid tiers are named by the size of your machine, from $30 a month to $200 a month, with a $10 monthly platform fee on the larger two, checked October 2026 (Vellum pricing). There is no shared repo where several people review what an agent may do.
• Work lands where you are. Vellum acts across your files and apps directly, gated by your own approvals. There is no per-task branch and no change request for a teammate to read before the work takes effect.
• Trust is set per person. Its permission model resolves one actor's trust level and enforces it. A company gets no roles, groups or per-resource permissions to run, and no audit trail to hand a reviewer.
• Governance is left to you. A team would have to assemble review, role separation and audit around it.

None of this is a flaw in the product Vellum set out to build. It is a personal assistant, and it is good at being one.

A company that lives in one repo

Kortix is the open-source AI Operating System: a company's agents, skills, memory and connectors live as text in one git repo the team owns (Kortix docs). Anyone can read the files, diff a change and roll one back.

acme-co/
  agents/       who does the work
  skills/       how the company does a job
  memory/       what it has learned so far
  kortix.yaml   connectors, triggers, the machine

An agent is a markdown file with a scoped reach into tools. A skill is markdown plus scripts that encode how the company does one job. Connector rules say allow, ask or block for each call, down to its arguments, and the connectors reach 3,000+ apps plus any MCP, OpenAPI, GraphQL or plain HTTP endpoint (Kortix on GitHub).

Where Vellum's identity is something one person shapes, Kortix's is a repository the whole team edits. A new hire clones it and starts with the same context. A change to how a job runs shows up as a diff anyone can review, and a mistake rolls back like any other commit.

Every session gets its own computer, every change gets a person

Kortix gives each session its own isolated cloud computer and its own branch. The agent can install, run and break anything, and only what it commits survives (Kortix docs).

The work does not merge itself. It reaches the main branch as a change request that a person reads as a diff and approves, and merge is default-deny for agents. Permissions are set per resource for people and agents, with roles, groups and an audit trail, and approval gates stay off until you set them. Kortix is free to self-host, and the team plan is $40 per seat per month with pooled credits (Kortix pricing).

Where Vellum is the better pick

For one person who wants an assistant that lives on their Mac, learns them and follows them to their phone, Vellum is the lighter answer. The identity layer is the point: you name it, you shape how it talks, and it remembers you without a team deciding anything. Vellum meets you on a phone, by voice and in Telegram, while Kortix's chat channels are Slack, Microsoft Teams and email. A single download asks nothing about projects or seats, and for work that is yours alone that is the right shape.

Side by side

When to pick which

If you are still weighing a personal agent against a company system, personal AI agents vs a company OS draws the same line across OpenClaw and Hermes, and Kortix vs OpenClaw makes the single-operator case in more depth.

Vellum is a good assistant for one person. When the work has to be owned and reviewed by a team, start with Kortix. Free to start, free to self-host. Try Kortix.