All AgentFleet features
Every feature in one list: what it does, whether it is on by default, and the command to turn it on, off or configure it. Each row links to the matching section of the docs when there is one.
On means it is active after a normal install. Off and Opt-in mean you turn it on. Automatic means it works without a setting. On request means you run it when you need it. Restart Claude Code after you change a setting. Commands starting with / run inside Claude Code; the others run in a terminal.
Lanes and routing#
Lanes are the subagents Claude hands work to. These features decide which model each lane uses and what happens when a model fails.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| Role-named lanes | AgentFleet installs 19 subagents named for a role, such as fleet-implement and fleet-review, never for a model. The names stay the same when you change provider. Lanes in the docs. | On After install. | Rebuild agentfleet syncRemove agentfleet uninstall |
| Capability tiers and ranking | Each model your provider offers gets a tier: cheap, fast, balanced or deep. Each lane takes the best live model for its tier and falls back to the next one. Tiers in the docs. | Automatic | Configure claude-fleet-setup --tier MODEL=TIERRefresh claude-fleet-setup --reconcile |
| Lane model preferences | Pin a lane to models you choose, keep models out of every lane, or reserve models for the lanes that name them. Tuning in the docs. | Opt-in No pins until you set them. | Pin claude-fleet-setup --prefer LANE=MODELUnpin claude-fleet-setup --prefer LANE=Exclude claude-fleet-setup --exclude MODELUndo claude-fleet-setup --include MODEL |
| Lane effort | Sets how much reasoning a lane uses: low, medium, high, xhigh or max. AgentFleet sets a default for each lane. Lane effort in the docs. | Automatic | Configure claude-fleet-setup --effort LANE=LEVELReset claude-fleet-setup --effort LANE= |
| Custom lanes | Add your own fleet-* lane by editing ~/.claude/fleet.json. Custom lanes survive updates and profile switches. Custom lanes in the docs. | Opt-in | Add Edit ~/.claude/fleet.json, then run claude-fleet-setup --reconcileGuided /fleet-setup |
| Project lane overrides | Pin a lane to a model for one project only. The pin lives in that project and does not change your other projects. Project lane overrides in the docs. | Opt-in | Pin agentfleet project pin LANE MODELUnpin agentfleet project unpin LANEShow agentfleet project show |
| Outage failover | When a gateway fails with a server error, AgentFleet marks the lane as failing for 15 minutes. Routing avoids it and tells Claude which lane to use instead. This is part of the lane-guard mod. Outages and lane reports in the docs. | On Runs with the lane-guard mod. | Off claude plugin disable lane-guard@agentfleetOn claude plugin enable lane-guard@agentfleet |
Work style#
These settings change how Claude approaches your tasks. Restart Claude Code after you change one.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| Work style | Sets how readily Claude hands work to a lane. balanced uses lanes for independent parts and risky reviews. direct uses a lane only when you ask. fleet delegates by default. Work style in the docs. | On Default: balanced. | Configure agentfleet config work-style direct|balanced|fleet |
| Reviews, fixes and retries | Claude reviews each lane's changes itself. It sends larger findings back to the lane that wrote the code, and retries a failed lane once. This works in every style that uses lanes. Reviews, fixes and retries in the docs. | Automatic No setting. | Change No command. Pick a work style instead. |
| Smallest complete change | New in 2.14.0 Claude and the implementation lanes find every place a change must reach, reuse existing code before adding any, and add nothing you did not ask for, while still finishing every part the task needs. Review lanes list code a change did not need. Smallest complete change in the docs. | Automatic No setting. | Change No command. Your project's CLAUDE.md can ask for more. |
| Ask first | For large, unclear or risky tasks, Claude agrees on an approach with you before it starts. Small, clear tasks start straight away. Ask first in the docs. | On | On agentfleet config ask-first onOff agentfleet config ask-first off |
| Branch check | Before Claude changes files in a git repository, it checks the branch and asks whether to continue on it, switch, or start a new one. It never switches or pulls without your answer. Branch check in the docs. | On | On agentfleet config branch-check onOff agentfleet config branch-check off |
| Reviewer access | Sets which tools the review lanes get: full (read, Bash, Edit, Write), run (read and Bash) or read-only. Reviewers edit only in an explicit FIX MODE. Reviewer modes in the docs. | On Default: full. | Configure agentfleet config reviewer-access full|run|read-only |
Context and compaction#
These features keep the context window from filling up, or show you how full it is.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| Auto-compact percentage | One percentage of usable context for every model. Claude compacts when the context reaches it. Choose 20 to 95, or off. Auto-compact in the docs. | On Default: 70%. | Configure agentfleet config compact 20-95Off agentfleet config compact off |
| token-weather | A forecast of how full the context window is, drawn above the prompt. It is a sample mod from Anthropic. token-weather in the docs. | On On since 2.5.0. | Off claude plugin disable token-weather@agentfleetOn claude plugin enable token-weather@agentfleetAsk agentfleet configure |
| session-info | One row above the prompt: the current model and effort, uptime, running subagents, git branch, prompts sent and cost at list price. session-info in the docs. | On On since 2.13.0. | Off claude plugin disable session-info@agentfleetOn claude plugin enable session-info@agentfleetAsk agentfleet configure |
| JEV (fast-jev-compaction) | A mod that replaces Claude Code's compaction summary. A fast Jev model decides which old tool calls to drop. It needs a paid TypeSafe API key and sends main-conversation text to api.typesafe.ai when it compacts. Optional mods in the docs. | Opt-in Not installed by default. | On agentfleet jev onOff agentfleet jev offCheck agentfleet jev status |
| Context Mode | An upstream plugin that indexes large tool output on your machine and returns only the excerpts Claude needs. Lanes may use only its search and stats tools. AgentFleet has not measured total-token savings. Context Mode in the docs. | Off Opt-in pilot. | On agentfleet config context-mode onOff agentfleet config context-mode offCheck agentfleet config context-mode status |
Mods#
Mods are plugins that run inside Claude Code and can draw in its interface. They need Claude Code 2.1.287 or later. They show in the terminal and the Code tab of the desktop app. token-weather and JEV are mods too; you find them under Context and compaction.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| Mods catalog | AgentFleet installs a local plugin catalog named agentfleet. It holds fleet-status, lane-guard, token-weather, session-info and brain, which are on by default, and fast-jev-compaction, which is opt-in. Mods in the docs. | On Turn all mods off at install time. | Manage /pluginSkip AGENTFLEET_MODS=0Or flag --no-mods |
| fleet-status (status band) | A band above the prompt that appears only when something needs action: a restart after an update, a failing lane, a usage warning above 80%, or a new suggestion. /fleet-status prints your version, profile and settings. /fleet-tasks opens a pane of this session's lanes. fleet-status in the docs. | On | Off claude plugin disable fleet-status@agentfleetOn claude plugin enable fleet-status@agentfleetOpen /fleet-statusOpen /fleet-tasks |
| lane-guard | Blocks covered git changes, GitHub writes, publishing, deployments, migrations, destructive SQL and dangerous deletions in fleet subagents. It checks your main conversation only with the main-session guard on. Turning it off also turns off outage failover. Lane guard in the docs. | On | Off claude plugin disable lane-guard@agentfleetOn claude plugin enable lane-guard@agentfleet |
| Main-session guard | Asks you before the main session, or a subagent that is not a fleet lane, discards work or data, or publishes: a force push, git reset --hard, a deploy, a migration, destructive SQL or a dangerous rm. Commits, pushes and pull requests run without a question. For sessions that skip permission prompts. Needs lane-guard. New in 2.15.0. Main-session guard in the docs. | Off | On agentfleet config main-session-guard onOff agentfleet config main-session-guard off |
| /brain | Opens a scrollable pane with agent-brain token usage, this week's estimated cost, and your project's memories when memory is on. brain in the docs. | On On since 2.7.0. | Open /brainOff claude plugin disable brain@agentfleetAsk agentfleet configure |
Usage, cost and reports#
See what your lanes and your main conversation use, and what it costs.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| Usage report | Summarizes recent lane runs and main-conversation tokens per model. For each lane it shows runs, outcomes, errors, forbidden commands and tokens. Usage and estimated cost in the docs. | Automatic Recorded locally. | Run agentfleet reportPeriod agentfleet report --days NScript agentfleet report --json |
| Prices and estimated cost | With prices set, the report and /brain show an estimated cost. Prices are in dollars per million tokens. A gateway that publishes prices in its model list is priced automatically. Usage and estimated cost in the docs. | Off Until prices are set. | Show agentfleet config pricesSet agentfleet config prices MODEL=INPUT,OUTPUT[,CACHE_READ,CACHE_WRITE]Remove agentfleet config prices off |
| Gateway usage | On a gateway profile, AgentFleet reads the spend and limits your gateway reports for your key (OmniRoute, OpenRouter or LiteLLM). The report, /brain and the status band show them. Reported by your gateway in the docs. | On Gateway profiles only. | On agentfleet config provider-usage onOff agentfleet config provider-usage off |
agent-brain#
agent-brain is a free tool that records the tokens and cost of each Claude Code session and can keep project memory on your machine. AgentFleet installs and connects it. It never asks you to sign in. See the agent-brain site for its own documentation.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| agent-brain usage tracking | Installs agent-brain from a checksum-verified release and connects it to Claude Code. With automatic updates on, AgentFleet also keeps it up to date. agent-brain in the docs. | Opt-in The wizard offers it; the default answer is yes. | On agentfleet config agent-brain onOff agentfleet config agent-brain off |
| agent-brain memory | Claude saves decisions and project context on your machine and reads them back in later sessions. Nothing leaves the machine. agent-brain in the docs. | Off Until you say yes. | On agentfleet config agent-brain-memory onOff agentfleet config agent-brain-memory off |
| Memory export and import | Moves a project's memories to another machine or to a colleague. Export writes a file. Import shows a summary and asks before it writes, and skips entries you already have. Memory export and import on the agent-brain site. | On request Needs agent-brain. | Export agent-brain memory export --format json --out FILEImport agent-brain memory import FILEPreview agent-brain memory import FILE --dry-run |
| Memory V2 upgrade | New in 2.12.0 AgentFleet upgrades agent-brain's memory to the current storage format, Memory V2, after it makes a backup copy. It never upgrades a shared PostgreSQL memory store automatically. Memory storage on the agent-brain site. | Automatic On install, update and | Run now agentfleet doctorPreview agent-brain memory upgrade-v2 |
Setup and maintenance#
Install AgentFleet, answer its questions, switch providers, check the install and keep it up to date.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| Installer | One command installs or updates a checksum-verified release. Run it again to update. It keeps your provider profile. Install in the docs. | On request | macOS, Linux curl -fsSL https://agentfleet.wecansync.com/install.sh | shWindows irm https://agentfleet.wecansync.com/install.ps1 | iex |
First-run wizard and agentfleet configure | New in 2.12.0 After an interactive install or update, the wizard asks only the questions you have not answered. agentfleet configure asks about every setting, grouped under headings, including branch check, Context Mode, gateway usage and the /brain mod. Enter keeps the current value. First-run wizard in the docs. | Automatic Runs after a terminal install or update. | Run agentfleet configureKeep all agentfleet configure --yesSkip AGENTFLEET_NO_CONFIGURE=1Show agentfleet config |
| Presets for teams | Apply a team's settings without questions, or export your own. Exports never contain secrets. Preferences in the docs. | Opt-in | Apply agentfleet configure --preset team.jsonExport agentfleet configure --export team.json |
| Provider profiles | Save the setup for each provider, and switch between them without reinstalling. Lanes keep their names. Use native for your Claude login. Profiles in the docs. | On Starts with one profile. | List agentfleet profilesSwitch agentfleet use NAMEAdd agentfleet add NAME --gateway-url URLSave agentfleet save NAMEDelete agentfleet remove NAME |
| /fleet-setup | An interactive wizard inside Claude Code. It audits the fleet, fixes stale model settings, tunes lanes and lets you add a provider. It never asks for a token in chat. fleet-setup in the docs. | On request | Run /fleet-setup |
| Doctor and status | Checks the install: models, settings, hooks, agents, skills, plugins and MCP health. It also runs the Memory V2 upgrade for agent-brain. Doctor in the docs. | On request | Verify agentfleet doctorAudit agentfleet status |
| Automatic updates | Once a day at Claude Code start, AgentFleet installs a newer release with the same major version in the background. A new major version is never installed automatically. Automatic updates in the docs. | On | On agentfleet auto-update onOff agentfleet auto-update offCheck agentfleet auto-update status |
| Update, roll back and uninstall | Update by hand, restore the settings from before the last install or update, or remove AgentFleet and restore your earlier settings. Upgrading in the docs. | On request | Update agentfleet updateRoll back agentfleet rollbackRemove agentfleet uninstall |
| Instruction check | Checks your Claude Code instruction files for stale lane names, duplicate lines and contradictions. It asks before it edits and backs up the file first. Instruction check in the docs. | On request The wizard offers it. | Check and edit agentfleet instructionsReport only agentfleet instructions --check |
Suggestions#
AgentFleet scans a project on your machine and suggests plugins, mods, hooks and tools that fit it. It installs only what you pick.
| Feature | What it does | Default | Commands |
|---|---|---|---|
| /fleet-suggest | Scans the project locally, makes no network requests and never reads .env values. It explains what each suggestion can access before it installs anything. Suggestions in the docs. | On request The status band shows one new suggestion at a time. | Run /fleet-suggestFrom a shell agentfleet suggestDismiss agentfleet suggest --dismiss ID |
| Usage learning | Counts command names, tools, documentation sites and slash commands for each project, never prompt text, to improve suggestions. Suggestions in the docs. | On On for new installs. | On agentfleet config learn onOff agentfleet config learn offDelete data agentfleet suggest --forget |
| Spec Kit | For a new or small project, /fleet-suggest offers Spec Kit. It writes a spec, a plan and a task list before code. Suggestions in the docs. | Opt-in Installed only if you pick it. | Offer /fleet-suggest |
| mcpvault | New in 2.12.0 An Obsidian vault tool that works without the Obsidian app. /fleet-suggest offers it when a project contains an Obsidian vault. Suggestions in the docs. | Opt-in Installed only if you pick it. | Offer /fleet-suggest |
| Graphify | New in 2.12.0 A code knowledge-graph tool. /fleet-suggest offers it for large repositories. Suggestions in the docs. | Opt-in Installed only if you pick it. | Offer /fleet-suggest |