Buy

Claude

Free & Premium

The Claude page manages your installation of Claude Code, Anthropic's coding agent for the terminal. You see running sessions, your plan limits and your usage, and you set up hooks, agents, MCP servers, plugins, settings and the CLAUDE.md files without editing JSON by hand. Every feature has its own tab.

Nothing on this page costs tokens. WebDonkey reads Claude's files and only runs claude commands that need no model, such as claude agents, claude auth status or claude plugin list.

What you do here

  • Set up Claude Code, sign in, see plan limits: Overview
  • See and end running sessions: Instances
  • Analyse tokens and API value per day, model and project (Premium): Usage
  • Group Claude work per client or job and export it as CSV (Premium): Projects to bill
  • Create and edit subagents: Agents
  • Run commands at fixed points in Claude's work: Hooks
  • Look up skills and slash commands: Skills & commands
  • Manage MCP servers for Claude Code: MCP
  • Switch plugins on and off: Plugins
  • Set model, permissions and environment variables: Settings
  • Read and edit CLAUDE.md and Claude's auto memory: Memory
  • Read past conversations and resume them (Premium): Sessions

Page layout

The tabs Overview, Instances, Usage, Agents, Hooks, Skills & commands, MCP, Plugins, Settings, Memory and Sessions sit in one row. If it does not fit the window, it scrolls sideways. On Free, Usage and Sessions carry a ★. The page remembers the tab you had open until you quit the app. ⌘K jumps straight into a tab, for example with Claude › Hooks.

If Claude Code is not installed, the page shows the setup instead of the tabs. If you are not signed in, the Almost done card with the sign-in sits above the tabs. The Overview covers both.

After a change to agents, hooks, MCP servers, plugins or settings, a notice confirms Saved. with the note Applies to newly started sessions. Running sessions only read their settings at startup.

What happens in the background

Before WebDonkey changes one of Claude's files, it puts a backup next to it, such as settings.json.wd-20261106-141500123. The last ten backups are kept. Unknown entries and their order are preserved. If Claude changed the file itself in the meantime, WebDonkey does not write, reloads it and asks you to make the change again. WebDonkey never writes ~/.claude.json directly, it changes MCP servers through claude mcp.

Free and Premium

Viewing is free everywhere: installation, instances, plan limits, hooks, agents, skills, MCP servers, plugins, settings and memory. You need Premium for:

  • the Usage and Sessions tabs,
  • creating, editing and deleting hooks and agents,
  • switching plugins on and off,
  • changing settings and CLAUDE.md files,
  • for MCP servers, everything beyond the free actions (see MCP).

With Hide Premium features under Settings › Navigation, the Usage and Sessions tabs disappear on Free.

WebDonkey does not need Claude Code. If it is missing, the page offers the setup, and everything else in WebDonkey works without it.

Overview Free

The Overview tab shows the state of your Claude Code installation at a glance: whether everything is installed and signed in, how much of your plan limits is used and what is running right now. If Claude Code is still missing, it guides you through setup and sign-in.

Where to find it

Claude, tab Overview. It opens first.

Setting up Claude Code

If Claude Code is not installed, the page shows the Set up Claude Code card.

  1. Click Set up Claude Code. The steps run one after another, the button reads Setting up ….
  2. Homebrew: the package manager for macOS, if it is still missing.
  3. Node.js and npm (MCP servers often run via npx).
  4. Claude Code itself.
  5. Available in the terminal: afterwards every new terminal finds the claude command.

WebDonkey detects what is already there and marks it already there. Nothing is installed twice, an existing installation through Anthropic's installer or an older one via npm is left alone. If a step fails, the reason appears below the card.

Signing in

After the installation the Almost done card appears. Signing in runs in the browser and cannot be automated.

  1. Click Open terminal and sign in. A terminal starts the sign-in, in WebDonkey's built-in terminal in the Sign in to Claude window.
  2. Follow the link in your browser and sign in.
  3. When the browser says you are signed in, close the terminal.

The page checks every 5 seconds whether the sign-in has arrived (Checking the status every 5 s …). After that everything works automatically.

The Claude Code card

Line Meaning
Installed … the version and where it comes from: Homebrew, npm or own installer
Signed in: … or Not signed in your account and, if known, your subscription
Available in the terminal a new terminal finds claude. Otherwise it reads Not found by a new terminal yet
npm available (for MCP servers via npx) without npm, MCP servers via npx will not start, and the line warns you

The arrow icon at the top right reads everything again.

Plan limits

Claude subscriptions have usage limits per time window. The Plan limits section shows them account-wide, across all sessions:

  • Week and, once Claude reports it, 5 hours; plus Spend limit for plans with one,
  • per bar the level in percent and the time of the reset. From 70 % the bar turns yellow, from 90 % red. Once the window is over, it reads reset,
  • below, where the value comes from and when it was measured: "from session “…” · 2 minutes ago" or "per /usage" when you checked your usage in Claude with /usage.

With several sessions running, the highest value per window counts, because usage can only grow within a window.

The statusline bridge

Claude reports the limits only to its statusline, the line at the bottom of the terminal. The statusline bridge passes them on to WebDonkey:

  1. Click Show live limits (install statusline bridge).
  2. WebDonkey sets wd claude statusline as the command for Claude's statusline. That command stores every update for WebDonkey and then runs your previous statusline command with the same input. Your statusline looks exactly as before.
  3. The limits appear with the next statusline update. Type something in any Claude session for that.

While the bridge is active, it reads Statusline bridge active with the Remove button. The confirmation names the statusline that comes back, or says that Claude shows none afterwards. If you set another statusline yourself in the meantime, it stays, and only WebDonkey's backup is removed. The bridge is only installed when you press the button, never on its own.

Right now

The tiles under Right now show what is going on. A click jumps to the matching tab:

  • instances, with the number of sessions working right now,
  • hooks, agents and MCP servers,
  • plugins, with the number switched on,
  • tokens today with their API value. With Premium, a click leads to the usage.

Tokens and API value are estimated from the local transcripts. With a subscription this is not what you pay.

Terminal and diagnosis

  • Open terminal with claude starts a new session in your home folder. In the built-in terminal, closing the window ends the session. The conversation stays and can be resumed.
  • Diagnose runs Claude's own diagnosis in the terminal.
  • Refresh reads the state again.

Where these terminals open is set under Settings › Open with › Terminal: in WebDonkey's built-in terminal, in the macOS Terminal or in WebDonkey Editor (see Open with).

Command line

wd claude shows installation, sign-in, the plan limits and the running instances. When the app is closed, it still names the last known plan limits. wd claude sessions lists the instances, wd claude usage --days 30 the usage (Premium). All of them understand --json. More in More tools.

See also

Instances Free

The Instances tab lists every running Claude session: interactive ones in a terminal as well as background sessions. You see what each one is doing, how full its context is, how much CPU and memory it uses and which app it runs in. From here you end a session that hangs or attach to a background session.

Where to find it

Claude, tab Instances. Or click the instances tile in the overview.

Columns

Column Meaning
Name the session's name, below it its folder, when it started and the process ID (PID). Background sessions carry background
Status idle, working, waiting for you, waiting or ended
Model · context the model, how full the context window is and what the session has cost so far. These values come from the statusline bridge, without it you see a dash
CPU · RAM CPU load and memory, including the MCP servers it started ("+3 child processes (MCP …)")
Runs in the app the session runs in, such as a terminal or an editor

The list comes from claude agents and refreshes every 5 seconds. When nothing runs, it reads "No Claude instance is running right now."

Ending a session

  1. Click the power icon in the row (End).
  2. Confirm. The conversation is kept either way, you continue it later with claude --resume or in the Sessions tab. Unsent input in that terminal is lost.
  3. If an interactive session does not react within 5 seconds, WebDonkey offers Force end. That ends it hard.

Background sessions

Claude Code can run sessions in the background. They get extra buttons:

Button What it does
Link icon (Attach) opens the session in a terminal. Closing only detaches, the session keeps running
Document icon (Output) shows what the session has printed
Power icon (Stop) stops the session. Its conversation is kept and can be continued with Attach
Bin (Remove) only for ended sessions: deletes the session and its worktree (the project's separate working copy). This cannot be undone

Where the terminal opens is set under Settings › Open with › Terminal (see Open with).

Troubleshooting

  • "The instance has already ended.": the session ended on its own in the meantime, the list updates shortly.
  • A session is missing: the list only shows what claude agents reports.
  • Model · context stays empty: install the statusline bridge in the Overview and type something in the session.

See also

Usage Premium

The Usage tab shows how many tokens Claude Code has used and what the same work would cost through the API: per day and model for the last 7, 30 or 90 days, plus the biggest projects and sessions. Below that you group folders into projects to bill. This is a Premium feature.

Where to find it

Claude, tab Usage. Or press ⌘K and type Claude › Usage.

How to do it

  1. Choose the period at the top: 7 days, 30 days or 90 days.
  2. Choose what the chart shows: Tokens or API value.
  3. Hover over a day. The tooltip lists every model and the Total.
  4. Below, see which projects and sessions used the most.

What you see

Section Content
Tokens per day / API value per day (estimated) stacked bars per day, one colour per model family. Above them the period's total, such as "12.4 M tokens · ≈ … API value in 30 days"
Projects to bill your projects, see Projects to bill
Top projects the folders with the most usage
Top sessions the sessions with the most usage
By model per model Input, Output, Cache write, Cache read and API value

Small items are grouped under Other. If there was no usage in the period, it reads "No Claude usage in this period."

How the numbers come about

All figures are estimated from Claude's local transcripts, the files in which Claude Code records every session. Their format is internal to Claude Code. WebDonkey reads them incrementally: a refresh only reads what is new.

The API value converts the tokens at Anthropic's list prices. With a Pro or Max subscription this is not your bill. It shows what the same work would cost through the API. Where Claude Code recorded a session's cost itself, as /cost shows it, that amount applies. The note below the chart says how much of it is for requests Claude Code does not write to the transcripts.

Where the prices come from

WebDonkey reads the prices from Anthropic's price page. The note below the chart names the source and the date, such as "Prices from Anthropic's price page, as of …". WebDonkey checks the page at startup and then once a day. That way new models and price changes are right without an app update. If the page cannot be read, the prices built into WebDonkey apply, and the note says so.

A new model without a published price shows Price follows instead of an amount. Its tokens already count. As soon as the price is known, WebDonkey also calculates the amounts retroactively.

Free and Premium

Without Premium the tab shows the lock Usage history with Premium. Projects to bill you already created stay visible above it and can be deleted. Today's token count is also shown on Free in the Overview.

Command line

wd claude usage --days 30 prints the usage per day, machine-readable with --json. This needs Premium. More in More tools.

See also

Projects to bill Premium

Projects to bill group the folders that belong to one client or job, such as a shop, its theme and the staging copy. WebDonkey counts how much Claude worked in those folders, per month or for any period, and exports the result as CSV. That gives you a solid basis for billing Claude effort to a client.

Where to find it

Claude, tab Usage, the Projects to bill section below the chart.

How to do it

  1. Click Project. The New project dialog opens.
  2. Enter a Name, for example the client or the job.
  3. Under Folders, add the folders with Add folder. Under From the usage history WebDonkey suggests folders from the usage report where Claude worked. One click takes one over.
  4. Choose a Brain if you like (see below).
  5. Click Save.
  6. At the top, choose Month and step through with the arrows, or choose Period and enter start and end.
  7. Click the project to expand it. Export CSV saves the bill as a file.

Fields in the dialog

Field Meaning
Name required, such as client or job
Folders at least one. Whatever Claude does in these folders and below counts for the project. Remove folder takes one out
Brain a notes folder from the Notes page. A session of this project that you start in WebDonkey Editor gets it as its brain. Default: No brain
From the usage history suggestions from your Claude history

Views

  • Projects: one row per project with API value, tokens and the number of sessions. Expanded, it shows the models with Input, Output, Cache write, Cache read and Equivalent API price, then By folder with the Of which agents column, a split per brain and every counted session with its Days in the period.
  • By brain: Claude usage in the period, split by the notes vaults sessions were started with. Sessions without a brain appear under Without brain.

Editing (pencil, Edit project) changes name, folders and brain. The bin (Delete project) asks first and removes only the project. The folders and Claude's history stay untouched.

How counting works

  • Every answer from Claude counts exactly once and for exactly one project: the one whose folder the work was done in. With nested folders the closest one counts, with the same folder in several projects the first.
  • An agent's answers count by the folder the agent worked in.
  • Work outside every project folder counts for the project the session was started in.
  • A session across a month boundary counts only with its days inside the period. Consecutive months therefore add up to exactly the total.
  • A session that WebDonkey Editor started in a project counts for that project as a whole.

The amounts equal the API list prices as of the date shown. With a subscription they are a reference value, not an actual bill from Anthropic. Tokens of models without a known price are not in the amounts, a note names them. Where the prices come from is explained under Usage.

CSV export

Export CSV asks where to save and suggests a file name from project and period. The file holds one line per session and model with From, To, session, folder, the token kinds and API value (USD), and the total at the end. If your locale uses the comma as decimal separator, WebDonkey writes the file so that Excel opens it correctly right away. The path then appears under Saved: ….

Free and Premium

Creating, analysing and exporting need Premium. Without Premium your projects stay visible above the tab's lock and can be deleted, but there are no figures.

Troubleshooting

What you see What to do
"Give the project a name." enter a name
"Add at least one folder." add a folder
"The start is after the end." swap start and end in the period
"No Claude usage in these folders in this period." check the period or the folders

See also

Agents Free & Premium

Agents (Claude Code also calls them subagents) are specialists Claude hands work to: one reviews code, one runs the tests, one knows your local setup. Each has its own instructions, its own tools and its own model. The Agents tab shows every agent that applies and creates new ones without you writing the Markdown files by hand.

Where to find it

Claude, tab Agents. Or press ⌘K and type Claude › Agents.

The list

Each card shows the name in its colour, its origin, the description, the model (or inherits model) and the tools (all tools or their number). The origin appears as a label: User (for all projects), Project with the project name or Plugin with the plugin's name. Clicking the name shows the whole file with instructions and path.

When there are none, it reads "No agents yet."

How to do it

  1. Click Create agent.
  2. Optionally pick a starting point under Start from:: Code reviewer, Test runner or WebDonkey helper. The templates are in English because agents work best with English instructions. You adapt them afterwards.
  3. Enter a Name in lower case, digits and dashes, such as code-reviewer.
  4. Choose Model and Colour.
  5. Write the Description: when should Claude use this agent? Claude decides from this text, so it is required.
  6. Leave All tools (inherit) on, or switch it off and pick the tools one by one.
  7. Choose under Applies to where the agent lives.
  8. Write the Instructions and click Save.

Fields and options

Field Meaning
Name file name and ID of the agent. Letters, digits, - and _ are allowed. While you type, capitals become lower case and spaces become dashes
Model inherit (the session's model) or sonnet, opus, haiku, fable
Colour colour of the name in Claude, or none
Description when Claude uses the agent
All tools (inherit) the agent may do everything the session may do. Switched off, you choose from Read, Edit, Write, MultiEdit, Bash, Grep, Glob, WebFetch, WebSearch, Task, TodoWrite and NotebookEdit
Applies to the User level (for you, in every project) or One project, shared with the team (Project). For Project you choose the Project below
Instructions the text the agent gets as its brief, in Markdown

When editing, entries the form does not know stay unchanged. The dialog lists them under Kept as they are: …, and the card shows them as +1 and so on. Level and project of an existing agent cannot be changed.

What happens in the background

An agent is a Markdown file: for you in Claude's agents folder (usually ~/.claude/agents/), for a project under .claude/agents/ in the project folder. Agents from plugins are read-only. New and changed agents apply to newly started sessions.

The WebDonkey helper template brings instructions on how an agent drives WebDonkey through the wd command. For an agent that always stays current, use the Keep the WebDonkey agent up to date switch under Hooks.

Free and Premium

Viewing is free. Create agent, editing and deleting need Premium. On Free the button carries the Premium badge and is locked.

Troubleshooting

What you see What to do
"Names may contain letters, digits, - and _ only." remove special characters from the name
"That name is already taken on this level." choose another name
"A description is required …" fill in the description
the agent does not show up in a running session restart the session

See also

Hooks Free & Premium

Hooks are commands Claude runs at fixed points in its work: before it uses a tool, after it edited a file or when it waits for you. With them you format files automatically, get a notification as soon as Claude needs your approval, or hear a sound when it is done. The Hooks tab shows all hooks grouped by event and creates new ones from templates.

Where to find it

Claude, tab Hooks. Or press ⌘K and type Claude › Hooks.

How to do it

  1. Click Add hook.
  2. Pick a template:
    • Notify me in WebDonkey when Claude waits: every notification from Claude adds an entry to WebDonkey's bell and shows a macOS notification.
    • Format files after editing: after every edit or write, a formatter runs over the changed file. Choose prettier, php-cs-fixer (.php only) or rustfmt (.rs only) below.
    • Play a sound when Claude is done: at the end of every answer macOS plays a system sound.
    • Own command: any event, any command.
  3. Under Applies to, choose the level and, for project levels, the Project.
  4. Click Save. The hook applies to newly started sessions.

Own command

Field Meaning
Event when the hook runs, such as PreToolUse, PostToolUse, Notification, Stop or SessionStart. The field suggests every known event
Tools (matcher) only for events that support it: which tools or sources the hook reacts to, such as Edit. Separate several with a vertical bar
Command the shell command to run
Timeout (s) how long the command may run at most. Leave it empty for Claude's default

Applies to

Level Where the hook lives
User (for you, in every project) in your Claude settings, applies everywhere
One project, shared with the team (Project) in the project's .claude/settings.json, ends up in the Git repository
One project, only on this Mac (Local) in the project's .claude/settings.local.json, stays on your Mac

The list

Each group carries the event's name and below it a description of when it occurs. For every hook you see:

  • its origin: User, Project, Local, Managed (set by your organisation) or Plugin,
  • the matcher, if set, and the kind, when it is not a plain command,
  • costs tokens for hooks that call a model every time they fire,
  • warnings such as script not found, not executable or duplicate,
  • the command itself.

Editable hooks have a pencil and a bin. The pencil opens Edit hook with the matcher and the hook as JSON. Only valid JSON with a type is saved. Hooks from plugins and managed hooks are read-only.

If disableAllHooks is set in the settings, a note above the list says none of these hooks run. You change that in the Settings tab.

Keep the WebDonkey agent up to date

Above the list sits the Keep the WebDonkey agent up to date switch. Switch it on and WebDonkey adds a hook on your User level for the start of every session. It then writes the agent webdonkey-cli to ~/.claude/agents/webdonkey-cli.md, with every wd command of the installed WebDonkey version. That way Claude knows the new commands after each update. The hook prints nothing into the session, Claude only reads the full agent when it hands it a task. Switching it off removes the hook again.

Free and Premium

Viewing is free. Add hook, editing, deleting and the Keep the WebDonkey agent up to date switch need Premium.

Troubleshooting

What you see What to do
script not found check the path in the command
not executable make the script executable, for example with chmod +x
duplicate delete one of the two identical hooks
the hook does not run restart the session, hooks only apply to new sessions. Check whether disableAllHooks is set
"The file was changed elsewhere in the meantime …" Claude changed the file itself. It has been reloaded, make the change again

See also

Skills & commands Free

The Skills & commands tab shows which skills and slash commands Claude knows. A skill is a set of instructions Claude draws on by itself for matching tasks. A slash command is a template you call in a session with /name. The tab is for looking things up, read-only and free.

Where to find it

Claude, tab Skills & commands. Or press ⌘K and type Claude › Skills & commands.

What you see

The list has two sections, Skills and Slash commands, each with its count. Every entry shows:

  • the name,
  • its origin: User (for you in every project), Project with the project name or Plugin with the plugin's name,
  • the description Claude reads when choosing.

The folder icon on the right opens the folder of the skill or the command file. Where it opens is set under Settings › Open with (see Open with): in WebDonkey Editor, in your code editor or in Finder. There you edit the file in your editor.

If WebDonkey finds nothing, it reads "No skills or commands found."

New skills and commands

You do not create them here. Skills and commands are Markdown files in Claude's folders, for you usually under ~/.claude/skills/ and ~/.claude/commands/, for a project under .claude/ in the project folder. Many also come with a plugin. New files show up here when you open the tab again after about a minute. For an immediate refresh, use the arrow icon on the Claude Code card in the Overview.

See also

MCP Free & Premium

The MCP tab shows the MCP servers Claude Code can use. MCP (Model Context Protocol) is how Claude gets extra tools, such as access to a database, a browser or an issue tracker. The tab is the same view as the MCP page, narrowed to Claude Code. What you change here also shows up there, and the other way round.

Where to find it

Claude, tab MCP. All clients together are on the MCP page.

What you do here

  • Running: the MCP servers Claude Code has started right now. See Running.
  • Configured: the servers from Claude's settings, from the User level, from projects and from plugins, with their state, such as running, not running, waiting for approval or off. See Configured.
  • Clients: the files Claude Code reads its servers from, and what came out of reading them. See Clients.
  • Add server: register a new server for Claude Code. See Add a server.
  • Check: start a server once as a test, or Check through Claude Code. See Check.
  • Log in: for servers that require a sign-in. See Log in.
  • Stop, remove, edit, turn servers on and off: see Stop and clean up and Edit, copy, remove.

The number of servers appears in the overview on the MCP servers tile.

What happens in the background

Claude Code keeps its MCP servers in ~/.claude.json and in a project's .mcp.json file. Every running Claude session keeps rewriting ~/.claude.json, so WebDonkey never changes it directly but uses the claude mcp … commands. New or changed servers apply to newly started sessions.

Free and Premium

The same limits as on the MCP page apply: viewing, checking, stopping, removing and logging in are free, and you may add up to 2 servers. Everything beyond that needs Premium. The details are under MCP.

See also

Plugins Free & Premium

A Claude plugin is a package that extends Claude Code with skills, slash commands, agents, hooks and MCP servers in one go. The Plugins tab lists the installed plugins and switches them on or off, without you having to know the claude plugin commands.

Where to find it

Claude, tab Plugins. Or click the plugins tile in the overview.

What you see

Each row shows:

  • the plugin's name and, after the @, the marketplace it comes from,
  • the version and the level it is installed on (User, Project, Local),
  • when it was last updated, such as "updated 3 days ago",
  • the Details button and the on or off switch.

When there are no plugins, it reads "No plugins installed."

How to do it

  1. Find the plugin in the list.
  2. Click the switch. on switches it on, off switches it off.
  3. Restart the Claude sessions it should apply to. Changes apply to newly started sessions.

Details opens a window with the output of claude plugin details for this plugin.

What plugins bring along

Whatever a plugin that is switched on brings along appears in the other tabs with the Plugin label: under Skills & commands, Agents, Hooks and MCP. Switch the plugin off and those entries disappear. A plugin's contents are read-only there.

What happens in the background

WebDonkey uses the commands claude plugin list, claude plugin enable, claude plugin disable and claude plugin details. It never writes the plugins' files itself. You install new plugins in Claude Code itself, and they show up here afterwards.

Free and Premium

Viewing and Details are free. Switching on and off needs Premium. On Free the list says "Switching plugins on and off needs Premium.", and the switches are locked.

See also

Settings Free & Premium

Claude Code reads its settings from several files: your own, those of a project and possibly those of your organisation. What finally applies is not always easy to see. The Settings tab shows the effective settings with their origin and changes the most important ones without you editing settings.json by hand.

Where to find it

Claude, tab Settings. Or press ⌘K and type Claude › Settings.

How to do it

  1. At the top, under Effective settings for, choose Me (no project) or a project.
  2. Find the setting under Adjustable here and click the pencil.
  3. Under Write to, choose the level whose file the value goes into.
  4. Enter the value and click Save. The change applies to newly started sessions.

The view

Section Content
Files, highest precedence last every settings file that applies to the selection, with level and path. A missing one is marked not present, a broken one invalid JSON
Adjustable here the settings you can change here, with a plain-language name and the key from settings.json in small type below
Everything else all other keys with their value, read-only

Next to each value are the levels it comes from (User, Project, Local, Managed). When one level overrides another, you see something like "beats User". A missing value reads not set. Under Everything else, the statusline is marked "managed by the statusline bridge (Overview)" when the bridge is installed.

Adjustable here

Setting Key Input
Default model for new sessions model a model name, such as opus
How Claude asks for permission by default permissions.defaultMode default, acceptEdits, plan, auto, dontAsk or bypassPermissions
Tools and commands Claude may use without asking permissions.allow a list of rules, such as Bash(npm run test:*)
Tools and commands Claude must never use permissions.deny a list of rules
Tools and commands Claude always asks about permissions.ask a list of rules
Switch every hook off at once disableAllHooks true or false
Environment variable env.NAME name (letters, digits, underscore) and value. The value is hidden while you type

With bypassPermissions the dialog warns you: Claude then runs every tool without asking. That is only meant for isolated environments.

Claude merges the lists from all levels. In the dialog you only edit the list of the level shown under Write to. The X on an entry removes it, the plus adds the new one.

Remove on this level deletes the key from the file of the chosen level. If it is set on another level, it still applies.

Levels

Level File
User your Claude settings, usually ~/.claude/settings.json
Project .claude/settings.json in the project, shared with the team
Local .claude/settings.local.json in the project, stays on your Mac
Managed defaults from your organisation, read-only

Free and Premium

Viewing is free. Changing needs Premium. On Free, Adjustable here carries the Premium badge and the pencils are locked.

Troubleshooting

What you see What to do
"The file was changed elsewhere in the meantime …" Claude changed the file itself. It has been reloaded, make the change again
invalid JSON on a file repair the file in an editor
a change has no effect a higher level overrides it, see "beats …", or the session still runs with the old values

See also

Memory Free & Premium

Claude Code reads instructions from CLAUDE.md files at every start: your own for all projects and those of a project. On top of that comes the auto memory, where Claude itself keeps notes on a project. The Memory tab shows all of these files in one place, as a readable preview or in the editor.

Where to find it

Claude, tab Memory. Or press ⌘K and type Claude › Memory.

How to do it

  1. Choose a file on the left. At the top is For all projects with your own CLAUDE.md, below it one section per project Claude knows.
  2. On the right you see the file in the editor. Preview shows it as formatted Markdown, Edit switches back.
  3. Change the text and click Save. For a moment it reads Saved.

The list on the left

Element Meaning
For all projects your CLAUDE.md on the User level, applies in every session
One section per project the project's CLAUDE.md, the one under .claude/ if it exists, the CLAUDE.local.md (just for you) and the auto memory files
Size and age shown next to every existing file
Also show files that do not exist yet shows CLAUDE.md files that are still missing. Saving creates them

Auto memory files carry a brain icon and, in the editor, the auto memory label. The MEMORY.md file, the index, always comes first.

Insert WebDonkey briefing

Insert WebDonkey briefing appends a ready-made English text block to the end of the file. It tells an AI agent how to drive WebDonkey through the wd command: services, sites, logs, databases. Claude can then, for example, restart a service or read a log in that project by itself. The button is not offered for auto memory files.

What happens in the background

Before writing, WebDonkey puts a backup next to the file, the last ten are kept. If Claude changed the file itself in the meantime, WebDonkey does not write, reloads it and asks you to make the change again. A changed CLAUDE.md applies to newly started sessions.

Free and Premium

Reading and the preview are free. Editing, Insert WebDonkey briefing and Save need Premium. On Free the editor is read-only.

Troubleshooting

  • "No memory files.": Claude does not know a project yet and you have no CLAUDE.md of your own. Switch on Also show files that do not exist yet to create one.
  • A project is missing: the list shows the projects Claude has already worked in. Start Claude once in the project folder.

See also

Sessions Premium

The Sessions tab lists your past conversations with Claude Code, across all projects. Find an old session with the search, read it back as it ran, and resume it with one click. This is a Premium feature.

Where to find it

Claude, tab Sessions. Or press ⌘K and type Claude › Sessions.

How to do it

  1. Narrow the list to one project with All projects, or type a term into Search titles and prompts.
  2. Click Read on a session to read it back.
  3. Click Resume to keep working. A terminal opens in the session's folder and continues it with claude --resume. Where it opens is set under Settings › Open with › Terminal (see Open with).

The list

Each session shows:

  • the title. On hover you see your first prompt,
  • running when the session is active right now. Resume is locked then, you find the running session under Instances,
  • Claudian for a chat from Claudian, the Claude plugin for Obsidian,
  • the project, the Git branch, the last activity, the tokens with their API value and, if involved, the number of subagents, such as "3 subagents".

If WebDonkey finds nothing, it reads "No sessions found."

Reading

Read opens the conversation in a window: your prompts with their time, Claude's answers as text and its tool calls as short lines with the tool and a summary. More than three calls in a row are folded into one line Tool calls: …, a click expands it. In very long sessions the oldest messages are left out, and the top reads "Earlier messages are left out."

If the session lives in a notes vault, links in [[double brackets]] open the matching note.

Resuming

Resume starts Claude in the folder the session ran in and continues the same conversation. A chat from Claudian continues with Claudian's model, prompt and settings, just like in Obsidian. In the built-in terminal, closing the window ends the session, the conversation stays.

What happens in the background

The list comes from Claude's local transcripts, the files in which Claude Code records every session. WebDonkey only reads them, no tokens are spent. Tokens and amounts are estimates, as in the Usage tab.

Free and Premium

Without Premium the tab shows the lock Session list with Premium. With Hide Premium features it disappears on Free. Without Premium you resume interrupted sessions in the terminal with claude --resume.

See also

Cookie preferences

This website uses cookies to ensure the best experience possible. Your choice applies to webdonkey.io and the shop at shop.webdonkey.io. More information...

Technically required

Cookies required for this website to function:

Always on
WebDonkey
WebDonkey
Provider: BuI Hinsche GmbH, Neumarkt 7, 06712 Zeitz, Germany
Purpose: Your consent, colour scheme and language for website and shop, price switches; with wd_debug in the address, the measurement events in the developer console for troubleshooting
Cookies and storage: Cookie wd_consent on .webdonkey.io, shared with shop.webdonkey.io (12 months); cookies wd_theme and wd_lang on .webdonkey.io, shared with shop.webdonkey.io, only when you switch colour scheme or language yourself (365 days); localStorage wd:consent, wd:theme, wd:lang; sessionStorage wd:acc, wd:term, wd:debug (only with wd_debug in the address; until the tab closes)

Statistics

Shows us which pages and products people look at on webdonkey.io and in the shop, where visitors come from and how they get to a purchase (with order number, amount and products, without name or email address). Provider: Google. Transfer to the USA possible (EU-US Data Privacy Framework). Your choice applies to webdonkey.io and the shop together.

Google Tag Manager, Google Analytics 4
Google Tag Manager
Provider: Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland
Purpose: Loads the services you agreed to, here and in the shop. Sets no cookies of its own.
Cookies and storage: none
Privacy policy
Google Analytics 4
Provider: Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland
Purpose: Usage statistics: page views, clicks, scroll depth, sources, devices, load times; in the shop also basket, checkout, purchases, registration and sign-in
Cookies and storage: _ga, _ga_<ID> (2 years)
Privacy policy

Marketing

Measures whether our ads on Facebook, Instagram and TikTok lead to visits and downloads on webdonkey.io and to registrations and purchases in the shop, and builds audiences for ads. Providers: Meta and TikTok, with whom we are jointly responsible for this. With this consent Google may also set advertising cookies (_gcl_…). Transfer to the USA and other countries outside the EU possible. Your choice applies to webdonkey.io and the shop together.

Google Tag Manager, Meta Pixel (Facebook, Instagram), TikTok Pixel
Google Tag Manager
Provider: Google Ireland Limited, Gordon House, Barrow Street, Dublin 4, Ireland
Purpose: Loads the services you agreed to, here and in the shop. Sets no cookies of its own.
Cookies and storage: none
Privacy policy
Meta Pixel (Facebook, Instagram)
Provider: Meta Platforms Ireland Limited, Merrion Road, Dublin 4, D04 X2K5, Ireland
Purpose: Ad measurement and target groups on Facebook and Instagram, also in the shop (basket, checkout, purchases, registration)
Cookies and storage: _fbp (3 months), _fbc (3 months, only if you arrive through an ad), fr on facebook.com (3 months)
Privacy policy
TikTok Pixel
Provider: TikTok Technology Limited, 10 Earlsfort Terrace, Dublin, D02 T380, Ireland
Purpose: Ad measurement and target groups on TikTok, also in the shop (basket, checkout, purchases, registration)
Cookies and storage: _ttp, _tt_enable_cookie (13 months)
Privacy policy