AGENT PROFILES

Amp

Amp Frontier Corporation · Coding agents

Coding agent and development environment that works with several models in the terminal, in apps and on cloud machines (orbs).

Typical tasks

Availability in Europe
Added to the catalog on Coding agent and development environment that works with several models in the terminal, in apps and on cloud machines (orbs).Details
01

What it does

7 features

Amp · Coding agent with cloud environments

Product and access

Amp is operated by Amp Frontier Corporation. You install the CLI with a script from the website; no public repository was reviewed.

Core features

7 features from provider sources · checked on 5 November 2026. Availability and limits are directly below.

  • Orbs as cloud machines

    Each orb thread gets a fresh, isolated environment with your code, plugins and tools. The agent keeps working when your laptop is closed.

    Source 11
  • The Dial: four modes

    The modes low, medium, high and ultra bundle model, reasoning effort, system prompt and tools. The page advises choosing by task rather than model name.

    Source 13
  • CLI with execute mode

    amp -x runs a single prompt without an interactive session and prints the answer. For scripts and CI you set AMP_API_KEY.

    Source 9
  • Save and share threads

    Threads have a URL and can be private, visible to the workspace, to groups (Enterprise only) or to anyone with the link, and can be exported as Markdown.

    Source 10
  • Schedules and webhooks

    A thread can have a schedule; a plugin in an orb can register webhooks that wake the agent.

    Source 12
  • Subagents

    Task handles delegated work, Read Thread reads other threads. Oracle, Librarian and Finder can be installed as plugins on a model you choose.

    Source 13
Show all 7 features
  • MCP and AGENTS.md

    You configure MCP servers locally or as remote definitions; AGENTS.md files carry project and personal instructions.

    Source 17

Usable AI models

Open choice

You choose a mode from low to ultra, Amp decides on the model behind it, and plugins can set their own models. Model Routing lets you connect your own keys, ChatGPT or Grok subscriptions, OpenRouter and other gateways, or an OpenAI- or Anthropic-compatible endpoint. Enterprise admins can forbid personal connections; the docs do not mention local models, only Ollama Cloud.

Model families named

  • Anthropic (Claude)
  • OpenAI (GPT)
  • Google (Gemini)
  • xAI (Grok)
  • Other models can be connected freely
Model choice
Automatic or your choice
Your own API key
Yes
Router or custom endpoint
Yes
Local models
Not stated
Depends on
Admin approval

The rating covers the generally available main access. Documented · Sources accessed on

Amp: Modes & Models · Amp: Model Routing · Amp: The Dial

02

Does it fit my task?

2 practical tasks · 6 requirements and limits

A good fit if …

  • you want agents to keep working on cloud machines and follow them from phone, Mac or terminal
  • you want to share threads with your team and link other threads as context
  • you want to build the agent into scripts or CI through the CLI

Less suitable if …

  • you expect approvals before every tool call to be built in
  • you need documented EU data hosting
  • you want to inspect or self-host the source

Editorial assessment based on provider documentation. Not a product test.

Practical tasks

Editorial examples based on documented features. No product tests performed.

01

Change in an orb, taken over by sync

Your input
A project with AGENTS.md and a limited task that you start in an orb.
Possible result
The agent works in the orb, and amp sync mirrors the changes into your local checkout.
What to check
Check the diff and tests locally; keep an eye on orb minutes and credits.
02

Scripted task in CI

Your input
A script that calls amp -x with a prompt and AMP_API_KEY.
Possible result
The agent’s final message appears on standard output.
What to check
Check the output in the pipeline; without built-in approvals, keep permissions small.
EDITORIAL EXAMPLE

Have a change made in an orb and take the result into your local checkout with amp sync.

Your possible prompt

Read the project rules. Investigate [error], propose a small change, and run the relevant checks. Show me the diff and any remaining uncertainties.

Adjust the task, data and permissions.

Requirements and limits

  • Requirement

    Account needed

    You create an account and sign in; you can optionally connect a ChatGPT Plus or Pro subscription.

    Source 2
  • Limitation

    No approval before tool calls

    According to the tools page, Amp does not ask for approval before running tools. Control is through a custom plugin; for untrusted sources the page suggests a policy or an isolated environment.

    Source 14
  • Limitation

    Webhooks: no signature check

    Amp does not verify signatures from external services; the plugin must do it. Requests are capped at 1 MB and the handler has 30 seconds.

    Source 12
  • Limitation

    Orb start limit

    Each user can start a burst of 20 metered orbs, then one every five minutes.

    Source 11
  • Not yet confirmed

    Data location in Europe

    The security page names Google Cloud in the US as primary infrastructure and says nothing about the EU or GDPR.

    Source 6
  • Not yet confirmed

    Plan coverage of schedules

    The orb automations page does not say which plans include schedules and webhooks.

    Source 12
03

Can I use it?

Availability still unclear

Europe & regional availability

Security page and terms reviewed. They name no data region or country release for Europe.

Reviewed scope: Cloud service with CLI and apps
Reviewed on:

  • Availability still unclear

    Europe: no provider statement

    The security page names Google Cloud in the US as primary infrastructure and mentions neither the EU nor GDPR; it points to security@ampcode.com. In the terms, GDPR appears only in the definition of sensitive data, and there is no statement on storage locations.

    Region, reason and source

    Region covered: EU/EEA, United Kingdom and Switzerland

    Documented reason: The reviewed sources name no country release for this.

    Official source · ampcode.com · Official source · ampcode.com

Europe and regulation explained

Accounts & permissions

Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.

Configuration covered: Amp account, optionally with a ChatGPT or Claude subscription

Sources reviewed:

Show all 3 points
  • Documented

    Account used

    You create an account and sign in to the CLI with amp login. For scripts you set AMP_API_KEY, and workspaces have service accounts. Teams lists SAML/OIDC SSO.

    Evidence · Evidence · Evidence · Evidence
  • Documented with conditions

    Admin approval

    Workspace MCP servers from .amp/settings.json must be approved before they run. Workspace admins can distribute permission plugins globally. Further enterprise rules (model allowlist, MCP registry) were not reviewed.

    Evidence · Evidence
  • Documented

    Data access & team sharing

    According to the tools page, Amp does not ask before tool calls. Orb threads run in a fresh, isolated environment.

    Evidence · Evidence
04

Memory and conversation continuity

3 of 6 criteria documented

Threads have a URL and can be shared, searched, referenced and deleted; AGENTS.md carries project knowledge. The pages describe no automatic long-term memory.

Threads and AGENTS.md

Sources accessed:

Documentation by criterion

StoragePartly documented

Storage and scope

Each thread has a URL on ampcode.com that opens in web, CLI and apps. The threads page does not say whether it is stored on a server or locally; the security page describes encryption for data stored by the Amp server. AGENTS.md lives as a file with you.

SelectionPartly documented

Selection and timing

Amp can offer to generate an AGENTS.md and you can ask it to update the file. Handoff starts a fresh thread with the relevant context. The reviewed pages describe no automatic remembering.

Edit & deleteDocumented

Inspect, edit and delete

Archiving removes a thread from lists while the URL still works. Deleting removes it permanently with its artifacts; per the security page, deleted threads are removed within 30 days. Export as Markdown by adding .md to the URL.

Access boundariesDocumented

Agent and channel boundaries

Visibility: private (plus workspace admins through audited access), workspace, group (Enterprise only) or anyone with the link. AGENTS.md is read from directories up to $HOME, subtrees, and personal and system-wide files.

RetrievalDocumented

Retrieval

With @T-… or @@ in the CLI you point to other threads, which the agent reads. The activity page and amp threads search look through visible threads.

New conversationsPartly documented

Session reset and compaction

You continue long threads in a new thread through handoff. The models page names context compaction as a task of supporting models; the threads page does not cover it.

05

Cost and effort

5 plans · Prices as of 5 November 2026

Prices and included usage

Pricing checked

Hobby is free, Individual costs 20 US dollars a month as Megawatt with 45,000 orb minutes and 200 US dollars as Gigawatt with 480,000. Teams cost nothing extra, Enterprise is custom.

PlanPriceIncluded
HobbyFree

All product features, pay-as-you-go for Amp-hosted orbs, free self-hosted orbs and your own runners.

Price source
Individual (Megawatt)20 US dollars per month

45,000 minutes of orb time for Amp-hosted orbs plus unlimited self-hosted orbs and repositories.

Price source
Individual (Gigawatt)200 US dollars per month

A larger allowance, according to the pricing page 480,000 orb minutes.

Price source
TeamsNo extra charge

Shared billing and pooled credits, shared portals and plugins, SAML/OIDC SSO; members pay per their own plan.

Price source
EnterpriseCustom

Pooled credits only with no paid user seats, SCIM sync, spend limits, audit logs and IP allowlisting.

Price source
  • Included usage does not roll over at the end of the month; purchased credits expire after twelve months. Orb usage is billed separately by the minute.

Pricing details

Setup

Create an account, install Amp by script or Homebrew and sign in with amp login. Orbs can run in the Amp cloud.

Operation & oversight

Based on public provider documentation. Assessments are editorial inferences from the stated properties. Actual time requirements and reliability have not been measured.

Deployment covered: CLI, orbs and apps

Sources reviewed:

Editorial assessment

The agent gets a lot of freedom, but there are no built-in approvals. If you need control, build it through plugins or work in isolated orbs.

Show all 6 points
  • Documented

    Files and working state

    An orb pauses after the run and wakes on the next prompt with conversation, files and services in place. A sleeping orb costs nothing.

    Evidence · Evidence
  • Documented

    Sign-ins

    amp login saves the account; AMP_API_KEY overrides saved accounts. For scripts the CLI rejects the short-lived session token from amp login; an access token from settings is used instead.

    Evidence · Evidence
  • Documented

    Starting further runs

    With -x a single prompt runs non-interactively and the CLI exits; with -ox it runs in an orb. Schedules (one per thread, once or repeating) and webhooks start the agent without you.

    Evidence · Evidence
  • Documented

    Human intervention

    Amp does not ask before tool calls. Remote MCP tools can be set to Ask First, then the owner approves each call. A snoozed thread that waits for approval or an answer shows in the sidebar and notifies you.

    Evidence · Evidence · Evidence
  • Documented

    Detecting errors

    If a scheduled run fails, Amp shows the error. Webhook events are retried after 5 seconds, with the wait growing up to 5 minutes; after an hour of rejection they are dropped and logged.

    Evidence
  • Documented

    Resuming after failures

    You resume paused schedules after fixing the problem. Archived threads can be unarchived unless their project was deleted; deleted ones cannot.

    Evidence · Evidence
06

Evidence and open questions

17 sources · 3 open points

Open points

The reviewed sources contain no reliable information on these points. A gap says nothing about the product.

Europe: Availability still unclear

Requirements and limits: not yet confirmed

Sources and evidence

Website, manual, pricing, security and terms pages reviewed. Feature descriptions come from the provider; example tasks are editorial. Hands-on tests are still pending.

Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.

Read profile as Markdown

How we assess the sources

Provider

According to the terms, Amp Frontier Corporation, based in San Francisco, operates the service; California law applies.

Operator
Amp Frontier Corporation
Organization
Provider
Provider sourcesSource 1 Source 2 Reviewed
Changes to this profile 1
  • Coding agent and development environment that works with several models in the terminal, in apps and on cloud machines (orbs).

These notes describe editorial changes. Source and pricing review dates appear in their respective sections.