What it does
5 features
Meta Muse · Coding agent
Product and access
Meta’s dedicated coding interface. Muse Code uses Muse models; the personal Muse agent has a separate profile.
Core features
5 features from provider sources · checked on 13 October 2026. Availability and limits are listed below.
Coordinated coding agents
Workflows divide larger development tasks among subagents, with progress and controls in a shared view.
Source 2Messages between sessions
Sessions can locally pass notes and results to other sessions owned by the same user.
Source 2Muse Wonder and Wonder Full
Meta lists Wonder for Muse Code and the API; Wonder Full is available to Muse Code subscribers. Check model selection in your plan.
Source 3Rewind the conversation
After confirmation, rewind returns to suitable earlier points in the conversation.
Source 2Control sessions through the SDK
A TypeScript library starts local sessions and integrates tools and permission decisions into custom programs.
Source 2
Usable AI models
Fixed choice across providers
Muse Code is built for Meta's Muse models; you switch with --model or in the model picker, and Muse Wonder Full comes only with larger subscriptions. The changelog also mentions third-party models in the picker and Claude and GPT sessions through Meta's gateway, but the docs do not describe this. The only own key the docs describe is a Meta API key; availability may vary by region, and Muse Wonder is in preview.
Model families named
- Meta (Llama, Muse)
- Anthropic (Claude)
- OpenAI (GPT)
- Model choice
- You choose
- Your own API key
- Yes
- Router or custom endpoint
- Not stated
- Local models
- Not stated
- Depends on
- Plan, Region, Preview or beta
The rating covers the generally available main access. Partly documented · Sources accessed on
Meta: Muse Code · Muse Code: Changelog · Muse Code: Configuration and context · Muse Code: Subscriptions · Muse Code: Authentication and billing
Does it fit my task?
2 practical tasks · 3 requirements and limits
A good fit if …
- you want a terminal coding agent on macOS or Windows with parallel subagents
- you want to use a subscription; plans start at 5 USD a month according to the product page
- you want to drive sessions from your own code via the SDK and reload them later
Less suitable if …
- you do not want to check whether Contributor Services as the default suits your data
- you need scheduled or background operation without a terminal session
Editorial assessment based on provider documentation. Not a product test.
Practical tasks
Editorial examples based on documented features. No product tests performed.
Prepare a change for review
- Your input
- A local repository and a bounded change request.
- Possible result
- Code changes and traceable work steps.
- What to check
- Review the diff, tests and executed commands.
Prepare a recurring development workflow
- Your input
- A clear request and separate implementation and review tasks.
- Possible result
- A coordinated workflow that can be reused after review.
- What to check
- Check access, integration of changes and the result of each subtask.
Prepare a bounded code change and review the diff.
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
Check platform details before installing
The product page lists macOS and Windows; the release post also mentions Linux. Consult current installation documentation for your system.
Source 1Limitation
SDK in developer preview
The TypeScript SDK provides sessions, tools and permission control as an early developer release.
Source 2Requirement
Choose subscription data-use settings
According to the policy, coding subscriptions use Contributor Services by default. Standard Services require changing the relevant account setting.
Source 5
Can I use it?
With requirements
Europe & regional availability
Developer-service country conditions checked separately from the personal Muse agent.
Reviewed scope: Muse Code and Meta model access
Reviewed on:
With requirements
Geographic Use Policy and account access
The 12 October policy does not list the EU/EEA, United Kingdom or Switzerland as blanket-restricted territories. Access still applies only where Meta has enabled it. The policy alone therefore does not confirm purchasing availability for a specific account.
Region, reason and source
Region covered: EU/EEA, United Kingdom and Switzerland: check account access
Documented reason: Meta sets geographic access and trade restrictions. An AI Act or DMA-based exclusion of these regions is not established here.
Official source · dev.meta.ai · Official source · dev.meta.ai
Accounts & permissions
Based on provider documentation. Permissions depend on the account, integration and organization policies. Review the permissions actually requested before use.
Configuration covered: Muse Code as a local terminal installation with a Meta account and subscription
Sources reviewed:
Show all 3 points
Documented with conditions
Account used
On sign-in you are taken to your single sign-on provider. The product page does not describe how the Meta Model API account is set up.
EvidenceOpen in reviewed sources
Admin approval
The reviewed pages mention no organisation or admin controls. With the SDK, your own code can decide the agent's permission requests.
EvidenceDocumented with conditions
Data access & team sharing
According to the provider, messages between sessions travel over a Unix socket between processes on your machine, without the network. According to the usage policy, Coding Harness subscribers are on Contributor Services by default regardless of the model chosen and must change an account setting to use Standard Services.
Evidence · Evidence
Memory and conversation continuity
4 of 6 criteria documented
Markdown in three scopes is retrieved through an index and selective file access.
Local Muse Code memory
Sources accessed:
Documentation by criterion
StorageDocumented
Storage and scope
Memory consists of local Markdown; shared project files reside under .agents/memory, private project files outside the repository.
SelectionPartly documented
Selection and timing
The guide describes maintained notes and automatic context delivery; automatic fact extraction and write timing remain unspecified.
Edit & deleteDocumented
Inspect, edit and delete
MEMORY.md and topic files can be read, edited and deleted directly; repository files can be versioned collaboratively.
Access boundariesDocumented
Agent and channel boundaries
There is private project memory by default, shared repository memory and personal cross-project memory.
RetrievalDocumented
Retrieval
Session start loads the index and up to 48 file paths; files are read on demand, and observers can insert relevant notes.
New conversationsPartly documented
Session reset and compaction
New sessions reload the file index. Session logs can be disabled separately; a memory-specific compaction rule is not described.
Cost and effort
4 plans · Prices as of 13 October 2026
Prices and included usage
Pricing checked
Monthly Muse Code subscriptions according to the product page.
| Plan | Price |
|---|---|
| Everyday Usage | USD 5/month |
| High Usage | USD 15/month |
| Power Usage | USD 50/month |
| Maximum Usage | USD 200/month |
Included in all plans: Plan-dependent usage allowance
- USD list prices. Check actual Wonder Full allowances and regional purchasing availability in your account; API token prices are billed separately.
Setup
Install the CLI and sign in with a Meta Model API account
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: Muse Code in the terminal with subagents, workflows and the SDK
Sources reviewed:
Editorial assessment
You work interactively in the terminal and can steer and roll back runs. Approval behaviour and scheduled operation are not documented, so test them yourself first.
Show all 6 points
Open in reviewed sources
Sign-ins
The reviewed pages say nothing about how long the sign-in remains valid.
EvidenceDocumented with conditions
Starting further runs
You start tasks in the terminal; according to the product page, multiple agents coordinate on every task and reviewers work in the background. Background or scheduled operation of the agent is not mentioned in the reviewed pages.
EvidenceOpen in reviewed sources
Human intervention
The reviewed pages do not say whether commands and file edits need prior confirmation; they only mention that the SDK hands permission requests to your code.
EvidenceDocumented
Resuming after failures
In the /workflows view you can stop, restart or cancel runs. Pressing Esc twice rewinds the conversation after you confirm.
Evidence
Evidence and open questions
5 sources · 0 open points
Sources and evidence
Official product page, release post and model announcement checked on 13 October 2026. This profile does not include hands-on testing.
Editorial team: Michael Meierhoff · sources reviewed on . Provider information is not independent evidence of performance.
Provider
Meta develops the Muse models, coding agent and Meta Model API.
- Operator
- Meta
- Organization
- Technology company
- Website
- dev.meta.ai
Changes to this profile 1
Meta’s coding agent added with subagents, Muse Wonder and Wonder Full, plus documented subscription prices.
These notes describe editorial changes. Source and pricing review dates appear in their respective sections.