Claude Managed Agents gives developers Anthropic's agent harness and sandboxes as an API to build into their own products. type.com runs your team's own automations end to end, with built-in triggers, shared connections, and a thread for every run.
14-day free trial. Runs on Claude and Codex.
Updated
Claude Managed Agents is Anthropic's hosted agent runtime, in public beta on the Claude Platform. Your application creates agents and sessions through the Claude API, and Anthropic runs the agent loop and an isolated Linux container for each session. It runs Claude models only and bills tokens plus $0.08 per session-hour of running time. It is built for developers putting agents into their own products.
type.com is a managed agent platform for a team's own work. You describe a job or write it in TypeScript, start it from a schedule, Slack, GitHub, Linear, a webhook, or email, and type.com runs it with Claude or Codex on the team's shared connections. Every run lands in a thread the team can read and continue. Plans start at $50 a month, with AI usage metered at the provider's API rates.
What decides it
Both products run the agent loop and a sandbox for you. These are the choices around them.
Nothing. Triggers, connections, delivery to Slack or email, and run history are part of the product.
The application: how sessions start, custom tools, answers to paused tool calls, and the interface people use.
The whole team. Runs land in shared threads that anyone in the Space can read and continue.
Developers, through the API, SDKs, and the Claude Console. The session viewer is for developers and admins.
Claude or Codex, set for each Space, including team members' own Claude or ChatGPT subscriptions.
Claude models, billed to a Claude Platform API account.
What it is
Anthropic describes Claude Managed Agents as a suite of composable APIs for building and deploying cloud-hosted agents at scale. It provides the harness and infrastructure for running Claude as an autonomous agent: the agent loop runs on Anthropic's servers, and each session gets a fresh Linux container with eight built-in tools for shell commands, files, and web search. You can also keep the orchestration with Anthropic and run tools in a sandbox you host.
Your application sends events to a session and streams back results through Anthropic's client SDKs or REST API, and the Claude Console gives developers a visual editor for agents and a session viewer for debugging. Anthropic positions it for long-running and asynchronous work, and for products that manage agents per customer and sessions per user.
Triggers
In Claude Managed Agents, a session starts when your application calls the API. Scheduled deployments, in public beta, also start sessions from a cron expression down to the minute. Webhooks tell your systems when a session changes state; they don't start one. A Slack message, a GitHub event, or an email reaches the agent only through code you write.
type.com starts automations from schedules (every minute to once a month), Slack messages, GitHub and Linear events, webhooks, emails sent to a Space, and RSS items. Instruction automations need no code: you describe the job, run a private test, and enable it. When a step becomes a fixed rule, it can move into TypeScript code that hands the judgment to the managed agent.
Credentials
Claude Managed Agents stores credentials in vaults, usually one per end user. Code in the sandbox sees a placeholder, and the real secret is added only to outgoing requests to the hosts the credential allows. Your application obtains OAuth tokens and registers them in the vault.
type.com connects apps once, through OAuth sign-in, API keys, service accounts, or hosted MCP servers, and assigns each connection to the Spaces that need it. Each run gets a short-lived token, and automation code never holds a provider credential. Scheduled runs act as the person who created the automation, with the Space's connections.
Cost
Claude Managed Agents bills two things: model tokens at Anthropic's API rates, and $0.08 per session-hour, counted only while a session is running. Web searches inside a session cost $10 per 1,000. Session budgets set a hard spend ceiling, and workspace spend limits cap a month.
type.com plans start at $50 a month for 2 team members, and Enterprise is $250 a month for 20. Each plan includes AI usage, metered at the provider's published API rates, and admins get a notice at 80% of the allowance. Team members can connect their own Claude or ChatGPT subscriptions, and eligible requests run through them instead.
Full comparison
Grouped by the decisions a technical team makes when it picks where agents run.
A managed agent platform for a team's automations, with a web app, a desktop app, Slack, a CLI, and MCP.
A suite of APIs for building and deploying cloud-hosted Claude agents, in public beta.
Teams automating their own work, such as reports, triage, reviews, and handoffs.
Developers building agents into their own products, and long-running asynchronous tasks.
Describe the job in a Space, or write a TypeScript automation and deploy it with the type.com CLI.
Create agents, environments, and sessions with Anthropic's SDKs or REST API, or configure agents in the Claude Console.
Claude or Codex, the same managed agent that works in the Space's threads.
Anthropic's configurable harness, with prompt caching and compaction built in.
Claude and Codex models, set for each Space.
Claude models, Claude 4.5 and later.
An isolated Linux sandbox for shell commands, Git, the GitHub CLI, and scripts.
A fresh Linux container per session with eight built-in tools, or a sandbox you host.
TypeScript code does the fixed steps and hands judgment to the agent in the same automation. Code automations are in early access.
The agent runs its tools. Custom tools run in your application, which returns each result to the session.
Schedules, Slack, GitHub, Linear, webhooks, email, and RSS.
API calls from your application, plus cron schedules through scheduled deployments, in public beta.
A private test run in a thread only you can see. Code automations run fixtures on every deploy.
Iterate on an agent's configuration in the Claude Console before writing code.
A thread for every run with the agent's messages and tool calls, plus run history with outputs and errors.
Event history stored server-side, a session viewer in the Console, and a record of each scheduled run.
Each run gets a short-lived token. Automation code never holds a provider credential.
Vaults hold credentials. Sandbox code sees a placeholder, and the secret is added only to outgoing requests to allowed hosts.
OAuth sign-in, API keys, service accounts, hosted MCP servers, and custom APIs, each assigned to the Spaces that need it.
Remote MCP servers, up to 20 per agent, and custom tools. Your application obtains OAuth tokens and registers them in a vault.
Everyone in the Space, unless a thread is private.
Developers and admins in the Claude Console workspace.
Scheduled runs act as the automation's creator, with the Space's connections.
The credentials in the vault your application passes when it creates the session.
Plans from $50 a month for 2 team members, with AI usage metered at the provider's API rates.
Model tokens plus $0.08 per session-hour while a session is running.
Eligible requests can run on a team member's own Claude or ChatGPT subscription.
Billed to your Claude Platform API account.
AI cost by Space in the analytics dashboard, with notices at 80% and 100% of the plan's allowance.
Session budgets with a hard spend ceiling, and monthly workspace spend limits.
Claude Managed Agents is a strong runtime, and some teams should pick it. It is probably the better fit if:
Claude Managed Agents is an agent runtime that developers build applications on. The alternatives are other cloud runtimes you build on, such as Amazon Bedrock AgentCore and the OpenAI Agents API, or a managed agent platform like type.com that already includes the triggers, connections, and shared run history a team needs.
| Option | Best fit |
|---|---|
| type.com | Best when a team wants its own automations run for it: triggers, shared connections, and a thread for every run, with Claude or Codex as the agent. |
| Amazon Bedrock AgentCore | Fits developers on AWS who want any model, with either AWS's managed harness or their own framework, and separate services for runtime, tools, identity, and observability. |
| OpenAI Agents API | Fits developers on OpenAI models who want OpenAI to run a managed Codex harness, in public beta, with a sandbox OpenAI hosts or one they run. |
| Self-hosting the Claude Agent SDK | Fits engineering teams that want Claude's agent loop in their own containers and will run the hosting, scheduling, and credentials themselves. |
Summarized from each vendor's own documentation on October 15, 2026. Check them for current features and pricing.
FAQ
Yes, for automating a team's own work. Claude Managed Agents is an API for developers, who build the application around the agent: how sessions start, custom tools, and the interface people use. type.com is a managed agent platform with that application already built: triggers, shared connections, and a thread for every run, with Claude or Codex as the agent.
Yes. Claude and Codex are the managed agents in type.com, set for each Space. Team members can also connect their own Claude or ChatGPT subscriptions, and eligible requests run through them instead of the workspace's AI usage allowance.
Yes. Scheduled deployments, in public beta, start sessions from a cron expression down to the minute. Other triggers, such as a Slack message, a GitHub event, or an email, need code in your application that calls the API. type.com starts automations from schedules, Slack, GitHub, Linear, webhooks, email, and RSS without that code.
Claude Managed Agents bills model tokens plus $0.08 per session-hour while a session is running. type.com plans start at $50 a month for 2 team members and include AI usage metered at the provider's published API rates, and eligible requests can run on a team member's own Claude or ChatGPT subscription instead.
Routines are a Claude Code feature in research preview for paid Claude plans. A routine is a saved prompt, repositories, and connectors that runs on a schedule as often as hourly, from an API call, or on GitHub events, and it belongs to one person's account, not the team. Claude Managed Agents is the developer API for building agents into your own product. type.com automations belong to a team's Space, so teammates see and continue every run.
Not directly. Anthropic's Console and session viewer are for developers and admins, and a chat, Slack, or Teams interface is something your application builds. In type.com, teammates work with the managed agent in Space threads, in Slack, and by email.
Claude Managed Agents is an agent runtime that developers build applications on. The alternatives are other cloud runtimes you build on, such as Amazon Bedrock AgentCore and the OpenAI Agents API, or a managed agent platform like type.com that already includes the triggers, connections, and shared run history a team needs. type.com: Best when a team wants its own automations run for it: triggers, shared connections, and a thread for every run, with Claude or Codex as the agent. Amazon Bedrock AgentCore: Fits developers on AWS who want any model, with either AWS's managed harness or their own framework, and separate services for runtime, tools, identity, and observability. OpenAI Agents API: Fits developers on OpenAI models who want OpenAI to run a managed Codex harness, in public beta, with a sandbox OpenAI hosts or one they run. Self-hosting the Claude Agent SDK: Fits engineering teams that want Claude's agent loop in their own containers and will run the hosting, scheduling, and credentials themselves.
When you are building an agent into your own product for your own users, with sessions and credentials per end user, or when you want Anthropic to run the agent loop for long sessions, including against a sandbox you host yourself.
type.com runs Claude and Codex automations with built-in triggers, shared connections, and a thread for every run, with no application to build around them.
Get started for free14-day free trial. Plans for 2 to 60 team members. Bring your own Claude or Codex subscription.
