Features
Everything around the agents, in one place
The front page shows the idea. This page shows what it adds up to: seven walkthroughs, then the catalogue, each entry with a number where there is one and a link to the part of the user guide that covers it.
- Off by defaulta switch someone has to turn on first
- AWS onlydepends on the AWS editions
- Experimentala proof of concept that may change
In practice
Seven walkthroughs
Recorded and captured in the Console against a demo backend. Click a picture to see it full size.
Split a review across three agents Off by default
A Claude session splits a design review three ways and gives each part to a different agent.
- The Claude parent starts three child sessions: Codex on the data model and migrations, Antigravity on API compatibility, Muse Code on the UI.
- Each child works in its own worktree and is asked to report back once when it finishes.
- The parent pulls the three reports together. The fleet graph shows four lanes on a time axis, with an arrow for every start and message.
Starting sessions from sessions and messages between sessions are both off until you switch them on. A parent runs 3 children at once by default (1–10).

From an incident to a fix Experimental
Two assistants, two jobs: one reads the monitoring, the other starts the work.
- Ask the SRE assistant what is open on PagerDuty. It reads the incident, then the CloudWatch metrics and logs, and narrows down the cause.
- The SRE assistant is read-only: it cannot change anything or start a session. So you take the fix to the Fleet Operator.
- The Fleet Operator starts a Codex session in a new worktree with the fix as its task, and the session's report comes back to that conversation.
Alerts do not arrive in the chat on their own; you have to ask. PagerDuty, Grafana, CloudWatch and AWS are connected under Settings › Ops & monitoring.

From a ticket to a pull request, and its review
The issue tracker in the left pane gathers the work waiting for you.
- Save a query, and GitHub issues and pull requests, Jira issues and Bitbucket pull requests land in one list.
- Start a session from an issue. The branch name comes from your naming rules (
feature/45-empty-list, say), and the work happens in a new worktree. - Open a pull request: its details show CI, merge conflicts and reviews, and “Start review” checks out the PR's own head branch in a new worktree.
- When GitHub says it can be merged, press Merge in the same details and confirm. Agents never merge; you do.
CI and conflict marks sit on the list rows for GitHub pull requests; for Bitbucket they are in the details. A row that changed since you last looked gets a dot and a bold title.

Approve the plan, then commit, in the browser
Nothing in this loop needs a terminal.
- Started in plan mode, the agent proposes a plan first. Approve it in place, write comments on it, or have another session review it.
- It gets to work. The working copy's changes sit right beside the chat.
- Stage and commit yourself from the commit graph view: real git, per working copy.

When a limit stops the work, move it to another agent
A session that hits its usage limit does not have to wait for the reset.
- A band above the chat offers the move, with the agents that still have quota first. “Move to another agent…” is in the session menu too.
- Pick the agent. The dialog lists what goes along: uncommitted changes, commits not yet pushed, and follow-ups not yet sent.
- The new session starts in the same working copy and reads a handover file of the conversation first. No AI writes a summary, and the original session stops.
Claude Code, Codex, OpenCode, Copilot, Antigravity, Muse Code and Kiro can take a moved session. A session that is working, or waits on a question or a plan, has to finish or be answered first.

One memory, every agent Off by default
What one session learns, the next session of any kind already knows.
- Turn on Agent memory in Settings. Sessions of every kind get the memory tools, and Claude Code's own memory can be imported.
- A Claude Code session saves how to run the checkout tests. The next morning a Codex session reads it before it starts.
- Every memory shows who wrote it and from which session. Pin, edit or forget it, or revert a change, from the Console.
Memory tidy-up, where a model suggests fixes to memories, is a separate switch, also off by default. Your own Claude Code or Claude Desktop can use the memory through /mcp, and claude.ai through the memory-only /mcp/memory.

Answer from your phone
Away from the desk, the Console runs in your phone's browser.
- The sessions overview shows which session is waiting on you.
- Open it and answer the question on its card, as you would at a desk.
- It carries on where it stopped, and the list shows it done.
Or answer in the session's Slack or Discord thread, once the chat bridge is connected and “Reply to steer (two-way)” is on (it is off by default).

And overnight: scheduled runs and resuming after a usage limit are on the front page, in a day with Agent Fleet.
Catalogue
What it can do
Grouped by area. Every card links to the user guide, which is the authority on details and limits.
Sessions working together
A session can start other sessions and message them; the Fleet Operator starts and steers sessions for you.

-
Child sessions Off by default
A session starts up to 3 children at once by default (1–10), of any of the 9 agent kinds, each in a new worktree. One level deep: children cannot start their own. Each is asked to report back once.
-
Messages between sessions Off by default
There are four kinds: request, question, answer and FYI. A stopped recipient is resumed to receive it. Up to 16 KiB a message and 6 a minute per sender.
-
Fleet graph
One lane per session on a time axis, children under their parent, and an arrow for every start, handoff, instruction and message.
-
A second opinion on the plan
From a plan waiting for approval, “Review in another session” lets you pick a different agent and model to review it.
-
Fleet Operator
A built-in assistant that starts, instructs, stops and resumes sessions for you, and asks before starting anything that uses resources.
-
Session reports and Auto-pilot
A session the operator instructed reports back when it goes idle, stops at a question or a plan, or exits abnormally. The operator replies on its own, up to 10 times in a row without you by default (50 at most). “Auto-pilot”, off by default, also answers questions with the session's recommendation and takes plans through review to approval.
-
Watches
A session can watch a PR's CI, reviews and merge, a GitHub Actions or Bitbucket Pipelines run, a GitHub release or tag, another session going idle, or a timer, and is told when it changes. Its live watches are listed above the message box. With “Tell sessions about their PR's feedback” on (off by default), CI failures and requested changes reach the PR's session, starting it if it is stopped.
-
Handoff
Carry a conversation over to another agent. The operator drafts the summary, and you confirm it before anything starts.
-
Move to another agent
Carry on in another agent kind with no AI summary in between: the conversation goes into a handover file, the new agent starts in the same working copy, and the original session stops. A session stopped at its usage limit offers the move itself.
Agent memory
One memory that every agent kind shares, and the snapshots of what each CLI remembers on its own.
-
One memory for every agent Off by default
Every agent kind reads and writes the same memory through its MCP tools, with a user scope and a scope per project. Pinned memories come first in what agents read at the start of work. A search returns 20 hits by default, 50 at most.
-
A view of its own
The Memory view on the activity bar lists your memories by scope, with search, scope and type filters and the tidy-up's pending suggestions. Each memory opens in a pane of its own, like a file, where you read, edit, pin or forget it.
-
Every change, and its author
The Console lists each change with the agent and session that made it. Edit, pin or forget a memory, bring a forgotten one back, purge its past revisions, or revert a change. The list pages back to the very first change.
-
Memory tidy-up Off by default
From agents' reports that a memory is stale or wrong, and from lint findings, a model drafts small fixes while the workspace is idle. Nothing in memory changes until you approve.
-
Bring in what agents already know
Import Claude Code's auto-memory, Muse Code's memory or a memory bundle, and write memories back to Claude Code's own memory.
-
Use it from your own Claude Off by default
Claude Code or Claude Desktop on another machine reads the memory through
/mcp, and claude.ai through the memory-only/mcp/memory; each writes it if you allow that, with your MCP token or an OAuth sign-in where the operator has turned it on. A tenant administrator can block it. -
Memory snapshots
What Claude Code, Codex and Muse Code remember on their own is snapshotted a few minutes after an agent stops. You can roll it back, and undo the rollback.
-
Ask a teammate: the inbox
Send a message or a request to members of your tenant, with up to 5 sessions attached. One recipient takes a request, can start a session from it, records its pull request and branch, and marks it done. Up to 20 recipients and 30 sends an hour.
-
Hand a session to another member
From the session's menu or a handoff card in the chat, pass a session to someone it is shared with. When unpushed work would be lost, “Push, then send” pushes it first.
-
Limit board Off by default
Show teammates where you stand on usage limits: the state only, a band per quota, or the exact value. A value goes stale after 60 minutes, and a tenant administrator can switch the board off.
Reading and editing files
The file viewer opens what agents write and what your team keeps, without leaving the Console.

-
Code
Highlighting for about 30 languages, change bars on lines git sees as modified, and a minimap up to 10,000 lines.
-
Markdown and Mermaid
Mermaid diagrams render in place, front matter shows as a property list, and
path:linelinks open the file at that line. -
Marp slides
A Markdown file with
marp: trueopens as a slide deck. -
draw.io diagrams
Rendered with no network access. Viewing only: you can edit the XML, not the drawing.
-
PDF
Text selection, zoom, and Ctrl/⌘ F to find across every page. Japanese PDFs without embedded fonts render too.
-
Office, ODF and EPUB
Word, Excel, PowerPoint, OpenDocument and EPUB, converted to Markdown in your browser, up to 40 MB. A plain preview: layout, shapes and images are not reproduced.
-
Images and the gallery
PNG to AVIF and SVG, zoomable to 8×. A gallery pane collects pictures; a generated one's settings are one click away.
-
Editing, with AI proposals
Save conflicts are detected, and an AI's proposed change is shown as a diff before you apply it. A save is up to 2 MiB.
-
Open a shell from a folder
Right-click a folder to start a shell session in it, named after the folder.
-
Download a folder as zip
A zip holds up to 20,000 files and 512 MiB, and leaves out
.gitandnode_modules. -
Not supported (yet)
Math (KaTeX), CSV as a table, Jupyter notebooks, and video or audio. These open as text, or not at all.
-
One inbox for three trackers
GitHub issues and pull requests, Jira issues and Bitbucket pull requests in one list, refreshed about every 5 minutes. A row with new activity gets a dot and a bold title, and GitHub and Bitbucket rows show their comment count.
-
CI and conflicts at a glance
Open GitHub pull requests carry CI and conflict marks in the list; a pull request's details show CI, conflicts, reviews and diff size.
-
Merge from the details
A GitHub pull request that GitHub says can be merged gets a Merge button, with a confirmation naming base ← head and the method. Bitbucket pull requests have none.
-
Branch named for the ticket
Starting from an issue fills in the branch name and its base from your naming rules.
-
Review from a pull request
“Start review” checks out the pull request's own head branch in a new worktree.
-
Memo queue
Jot a note on your phone and send it to a session from your desk: memos belong to you, not to a device. An assistant can sort them.
-
Memo auto-run
Give a memo category an instruction and an agent, and its waiting memos run as sessions: by “Run now”, or by themselves in a time window or once enough are waiting. A memo a session adds waits for your approval. A tenant administrator allows it first.
-
Every run in one view
The Auto-run view lists the runs by category, with the work item, the pull request and its state, and the sessions each run started, such as reviewers. Stop a run's sessions or delete its worktree from its row, and clear finished runs at once.
-
In order, one part per run
A memo can wait until the memos it builds on have run and their pull requests are merged. Approving a memo also covers the follow-ups its run adds, up to 4 by default, so a big task lands one pull request per run. A run that cannot do the work yet hands its memo back for later.
-
Scheduled runs
Tell the Fleet Operator “every night at 2:00, …”. The schedule wakes a stopped workspace to run it, and can run only when a saved work item query has something new.
Git, in the Console
Real git per working copy and worktree, with the housekeeping that parallel work needs.

-
Commit graph
Branch lanes and diffs, staging and commit, and the graphs of submodules.
-
Worktrees against their parent
Each worktree shows where it stands against its parent (even, unmerged, fast-forwardable or diverged), with a one-click fast-forward.
-
Restore from the trash
Deleting a worktree keeps its commit, branch and uncommitted changes. Restoring puts it back at the same path, sessions included.
-
Built-in git hosting
Internal repositories with Git LFS, shared within your tenant. An administrator sets how many and how much LFS storage.
-
Submodules and LFS
Submodules are fetched with the clone, and SSH URLs are rewritten to HTTPS so they authenticate.
-
Git Flow
Initialize Git Flow for a repository from its row. It writes the configuration only, without committing or checking anything out.
-
Subversion
Check out SVN repositories too, with saved credentials. No worktrees or commit graph there.
A development environment around the agent
Databases, a browser and previews inside the workspace, with no Docker required.
-
Databases per working copy
Postgres or MySQL 8.4 for each working copy, with no Docker. The server stops after 30 idle minutes.
-
Browser pane
Your dev server in a real browser inside the Console, with HMR, WebSockets and cookies working. 2 per workspace, on desktop, tablet and phone. It is not available on ECS Fargate or the GKE preview.
-
Preview URLs
A URL per port that you can show to your tenant, or open without sign-in for a while. Issued only where the deployment provides preview subdomains.
-
Hand the browser to a person
When an agent reaches a one-time code, it hands you the page. You type the code and press “Action complete”. This is not available on ECS Fargate or the GKE preview.
-
A real terminal
A plain shell, or an agent started as Terminal (CLI), can be attached as a live PTY, with a control-key row on phones. Managed sessions show the chat only.
-
CLIs already signed in
gh,jiraandbktin every session sign in through your connections, so agents work with GitHub, Jira and Bitbucket without a token in the shell. Thejirawrapper retries reads on a flaky network, edits and deletes comments, and handles attachments. With Jira connected,confluenceworks through the same connection. -
Sign in with Google on a preview URL
An app that signs in with Google can sign in on its preview URL through a development identity provider instead, with test personas and copy-ready configuration. It is on by default wherever preview subdomains are issued.
-
Toolchains
The exact JDK (Temurin and Corretto) and Node.js patch in use, an update to the newest patch with no restart, and Uninstall for the versions you no longer need.
-
Your own assistants
Give an assistant a persona, an agent and model, and a tool permission: none, read or write access to Agent Fleet.
-
Translation without a turn
“Translate” reads an agent's answer in your language through one model call outside the session.
-
Reply suggestions
Chips learned from your own messages, with no AI involved. ✨ asks an AI, and spends tokens only when pressed.
-
A work plan that survives compaction
Compact a long assistant chat (it happens automatically above 90%) and the work plan is carried over word for word.
-
Branch from a message
Fork a conversation from one of your earlier messages, and try another direction without losing the first.
-
Switch the agent mid-conversation
Move an assistant chat to another agent or model and keep the conversation.
-
Past a usage limit
A chat turn that fails on a usage or rate limit before it shows anything is re-run once on the next usable agent in your priority list, with a notice. The conversation keeps its agent.
-
Slack and Discord Off by default
Each session gets its own thread. With “Reply to steer (two-way)” on, you answer there with buttons, and
!start <first instruction>starts a new session with it. -
Hold follow-ups while a turn runs
Queue follow-ups instead of sending them at once. They go out one turn at a time when the agent finishes, and you can edit, reorder or send one right away.
-
Export a conversation as Markdown
The session menu exports it at three levels: the conversation only, with tool names, or with tool output. Strings that look like secrets are masked.
Operations and SRE
Read your monitoring from a chat, and run AWS and Google Cloud commands as yourself.
-
SRE assistant Experimental
Reads PagerDuty, Grafana, CloudWatch and AWS. Read-only: it cannot start sessions, and alerts do not arrive on their own; you have to ask.
-
AWS tools, read-only first
The Agent Toolkit for AWS starts read-only. Write tools are a separate switch, and they can change real resources.
-
AWS commands run only in the account you name
af-aws-execruns a command as your profile and refuses unless it belongs to the account you name, so nothing falls back to the workspace's own role. -
SSM login
Reach in-house hosts through SSM Session Manager, signing in with IAM Identity Center. No AWS secrets are stored in Agent Fleet.
-
Google Cloud commands as yourself
af-gcloud-execruns gcloud,kubectlagainst GKE and Terraform as your own Google Cloud profile, with a short-lived token renewed in the background. No Google credential is stored.
MCP
Bring your own tools to the agents, and drive Agent Fleet from outside.
-
Your MCP servers
Register a remote server with no authentication, a Bearer token, custom headers or OAuth, or run a stdio server. Secrets are stored encrypted, and an OAuth token stays in the workspace behind a local relay, never in a CLI's configuration. Attach servers to sessions and to the assistants you create, from Settings › MCP servers (used by sessions).
-
Drive it from an external agent
Operate your workspace from Claude Code or Claude Desktop with an MCP token: read, write or admin scope, expiring after 90 days by default, revocable. Once the operator turns it on, claude.ai, ChatGPT, Claude Code and Codex can sign in with OAuth instead, with read or write access; such a connection ends after 30 days unused or 90 days after you allowed it.
-
Distribute to the tenant
A tenant administrator hands a remote MCP server to everyone in the tenant at once: with shared values, with a credential each member enters, or with an OAuth sign-in each member makes for themselves.
Image generation and your own engines
Generate pictures on an engine an administrator set up, or through your own Codex or Antigravity sign-in, and run open models inside the deployment.

-
The studio
Talk to an agent to refine the draft. It may trial the exact draft on screen, one picture at a time; you press the button for the batch. Every edit is in the history, and you can go back.
-
ComfyUI engines
Point the deployment at your own ComfyUI on the network, or run GPU engines on AWS. An administrator registers them.
-
A GPU only when it is needed AWS only
On
ecs-ec2, a GPU instance starts when a job asks for it and stops once it idles. The first image job waits about 3 minutes, and a chat engine about 9. -
llama.cpp sessions
Run open models such as Qwen3, GPT-OSS and Gemma as a session of their own, with no sign-in. A cold first turn takes minutes.
-
Pictures from any session Off by default
Let sessions call the image generator as a tool.
-
Command palette
Ctrl/⌘ P finds sessions, commands, changed files and files, and lists sessions waiting on you first. It also searches past conversations and, while Agent memory is on, the memory.
-
Leader key
Ctrl/⌘ K opens a command menu of short key sequences.
-
Next thing that needs you
Alt U jumps to the next session waiting on your answer, then to unread notifications.
-
Panes
Up to 4 × 2 panes side by side; any pane pops out into its own browser tab.
-
On a phone
Swipe left and right through the running sessions. A row of control keys (Esc, Tab, arrows, ^C) sits under the terminal.
-
Read aloud Off by default
Answers read out in a voice per session, placed left or right to match its pane. Zundamon (VOICEVOX, started by an administrator) or Amazon Polly.
Governance and administration
Rules that reach every agent, and controls an administrator can hold.
-
Three layers of instructions
The fleet policy from the operator and your own agent instructions reach each CLI in its own idiom; Cursor takes neither. The third layer is each repository's
CLAUDE.md/AGENTS.md, which the CLI reads itself (opencode reads onlyAGENTS.md). -
Your agents’ skills, read-only
Which skill folders each CLI reads, which copy wins when two share a name, and what each file says: personal skills in Settings › Skills, a repository’s from its row menu.
-
Member defaults
A tenant administrator names the team's agent and sets the model, execution method, permission prompts and budget the launch dialog opens with, and the team repositories listed first. A member's own choices win.
-
Models you don't use
Hide the models your plan bills extra for: they leave the pickers, and launches naming them are refused.
-
Which model routes members may use
A tenant administrator limits routes by type of credential and destination: no subscriptions, only Amazon Bedrock, or every metered key but one provider. Each layer can only narrow the one above, and what is forbidden disappears from the launch dialog, the agents and credentials pages, image generation and the assistant.
-
Audit log
Changes are recorded; terminal input and output are not stored.
-
Per-member limits
Concurrent sessions, and each member's workspace size: memory, CPU and disk.
-
Boost the next start
On ECS EC2, where the tenant allows it, a member raises the next start one size up in the same class for 1 to 24 hours, from Settings › Machine. The monthly hours budget is stored but not enforced yet.
-
Allowed networks
Limit a tenant to IPv4 and IPv6 ranges you list.
-
Outbound control
The deployment's operator records where workspaces connect and builds an allowlist from it; members request a host from Settings. Enforcing it holds a workspace only on AWS behind the optional egress gateway, a preview.
-
A tenant's own sign-in
A tenant can bring its own OIDC provider or GitHub organization, once the deployment's administrator approves it.
-
A member roster that flags trouble
Tenant administrators see members as cards with their state, home disk, sessions and last activity, and can filter for those who need attention, such as a home that is nearly full.
-
Tokens by feature, agent and model
Over 24 hours, 7 or 30 days. Calls that report no tokens count as unmeasured, never as free.
-
API-equivalent cost
Tokens priced at the published API list price. It is an estimate, not what a flat subscription bills you.
-
Plan limits on the top bar
How much of each plan's limit is used, for Claude, Codex, Copilot, Antigravity and Muse Code.
-
Resume after a limit
A turn cut short by a usage limit picks up again once the limit resets. This works for Claude, Muse Code, and Codex when Managed; any other session can move to an agent that still has quota. On a Claude subscription you choose who resumes: Agent Fleet (the workspace can stop while it waits), Claude Code itself, or nobody.
-
Cloud cost per workspace AWS only
The cost directly attributable to each workspace, about a day behind.
-
Running-time heatmap
Hours of the day against dates, coloured by time running or by session count.
-
Resource use by day AWS only
On ECS EC2, a table per UTC day and box: CPU, memory, the memory-pressure peak and minutes above the warning threshold, kept for 90 days. A tenant administrator sees it for each member and for the whole tenant.
-
A budget per session
Give a session a budget in US dollars. It stops after the turn that reaches it, and a turn that runs to 2× the budget is halted. The spend is an estimate at list price, and some agents report too little to be stopped.
Try it on your own machine
The native edition needs Linux on x86_64 or arm64, WSL2, or a Linux VM on an Apple Silicon Mac, and no Docker.



