Agenta vs Kortix

Agenta

• Teams that want to self-host their agent platform under an MIT license
• Paying by usage for a whole team instead of per seat
• Recurring jobs such as research briefs, code review and reports

Kortix

• Teams that want agent configuration versioned in a git repository
• Organizations that need to self-host or deploy in their own cloud
• Engineering-led teams comfortable reviewing agent work as diffs

Agenta

• AI model costs are not included: you connect your own provider credentials and pay the provider directly.
• The Hobby plan stops running agents after 5,000 runs in a month and allows 2 team members and 20 evaluations a month.
• On Agenta Cloud, SSO and role-based access control start at Business ($299 a month); audit logs and custom domains are Enterprise only.
• Running agents on a Claude or ChatGPT subscription works only on a self-hosted deployment, not on Agenta Cloud.

Kortix

• The Free plan is limited to 1 project and 200 sandbox credits a month, and you supply model access yourself.
• Kortix marks up managed model costs by 20% and most tool calls by 50%; with your own API key it charges a flat 10% platform fee instead, waived on Free.
• SAML SSO, SCIM, advanced RBAC and audit logs need Enterprise, and a self-hosted instance needs an Enterprise license to unlock them.
• iPhone, iPad and Android apps are listed as coming soon, and self-hosted instances still run agent sandboxes on an outside provider such as Daytona.

Agenta

Agents built by chatting

You describe the job in chat and refine the agent's instructions and skills as you go. Prebuilt templates cover engineering, support, sales, knowledge and operations.

Automations and triggers

An agent can run on a schedule or start when an event happens in a connected tool, such as a new pull request. Every run is logged with its duration and estimated cost.

Integrations, MCP and skills

Agents work across more than 1,000 integrations and any MCP server you add. Skills are reusable instruction files that can be swapped without rewriting the agent.

Harness and model choice

Three harnesses ship with it: Claude Code, Pi and Codex. Models come from your own provider accounts, including OpenAI, Anthropic, Gemini, Mistral, AWS Bedrock and Azure OpenAI.

Permissions, approvals and a secret vault

You decide which actions run automatically, which wait for approval and which are blocked. Credentials sit in a vault the agent can use but cannot read.

Tracing, evaluations and versions

Each run produces a trace of its model and tool calls, tokens and latency. Agent changes are versioned and can be compared, and evaluations include LLM-as-a-judge.

Kortix

Company as code

Agents, skills, memory, connector settings and triggers are text files in one git repository, so any change can be searched, diffed and rolled back.

A computer per session

Each session boots its own isolated Linux machine with your repo and tools already on it, and many sessions can run in parallel without affecting one another.

Change requests and tool rules

Session work lands on the main branch as a change request you read as a diff first, and each tool call can be set to allow, ask or block.

Any model, your keys

Pick a model per agent, session or message, using your own API key, a ChatGPT subscription, any OpenAI-compatible endpoint or Kortix's managed gateway.

Connectors

Connects to 3,000+ apps plus MCP servers, OpenAPI, GraphQL and raw HTTP. Connector credentials are brokered server-side and never enter the session's machine.

Channels and automations

Sessions start from the web app, Slack, Microsoft Teams, the CLI or the API, or with nobody present from cron schedules and signed webhooks. Email is in preview.

Agenta

• Hobby$0

  Agenta Cloud with 2 team members, unlimited projects, agents and workflows, 5,000 agent runs a month and 1-week trace retention.

• Open source$0

  Self-hosted edition with unlimited users, agents and runs, role-based access control, SSO, tracing and evaluations. You run it yourself with Docker Compose.

• Pro$29

  per month, any number of users

  Unlimited team members, schedules and event triggers, 10,000 agent runs a month, then $5 per additional 10,000, and 1-month trace retention.

• Business$299

  per month, any number of users

  Everything in Pro plus team roles and role-based access control, SSO, a SOC 2 Type II report, 3-month trace retention and priority support.

• EnterpriseCustom

  Cloud or self-hosted, with audit logs, custom domains, deployment support, dedicated support and a custom service-level agreement.

Kortix

• Free$0

  200 credits a month for sandbox compute, 1 project, and model access through your own API key or a connected ChatGPT subscription.

• Team$40

  per seat / month

  2,500 pooled credits a month per seat (about 125 default Agent Computer hours), up to 200 projects and 100 seats, credit top-ups and email support.

• EnterpriseCustom

  Everything in Team, plus SAML SSO and SCIM, advanced RBAC and audit logs, cloud, VPC or on-prem deployment, and an SLA, DPA and dedicated support.

Agenta

• Startups

  Discounted or free access for startups

  The pricing FAQ says early-stage startups can email the team to ask about discounted or free access, describing what they are building and how they plan to use Agenta.

• Education

  Discounts for students and researchers

  The same FAQ invites students and researchers to email the team about discounted or free access. No fixed rate is published.

Kortix

• Other savings

  Free to self-host

  Kortix is open source and can be self-hosted at no charge as one Docker Compose stack built from the same images as the managed cloud.