The best way to control Claude Code, Codex and more from any computer or phone

main
md 176 lines 9.1 KB
Rendered Raw
1 # Claude
2
3 T3 Code uses Claude Code's login and configuration. Start with the default provider
4 for one account; [provider setup](./install.md#providers) covers installation and
5 shared provider settings.
6
7 ## Separate accounts or configurations
8
9 Use a separate Claude config directory for each account. This also works for named
10 presets that need different Claude settings or a router connection.
11
12 Keep your existing account in the default directory. On the environment's machine,
13 create the second login:
14
15 ```bash
16 mkdir -p ~/.claude_personal
17 CLAUDE_CONFIG_DIR=~/.claude_personal claude auth login
18 ```
19
20 Add another Claude instance in **Settings > Providers**:
21
22 | Instance | Binary path | CLAUDE_CONFIG_DIR path |
23 | --------------- | ----------- | ---------------------- |
24 | Claude Work | `claude` | Leave empty |
25 | Claude Personal | `claude` | `~/.claude_personal` |
26
27 On mobile, tap **Add Claude account** in **Settings > Providers** and enter a name and the
28 config folder. The login itself still happens on the environment's machine.
29
30 An empty config-directory setting uses Claude Code's normal configuration. The
31 custom setting changes `CLAUDE_CONFIG_DIR`, leaving `HOME` and the system keychain
32 location intact. Use the same variable for the login command. Setting `HOME`
33 instead can put credentials where this provider will not find them.
34
35 Check the account reported in provider settings after signing in. Existing
36 threads can switch only between Claude instances with the same config directory.
37 Separate account directories stay isolated, including their local conversation
38 state. Claude does not have Codex's shared-home and shadow-home arrangement.
39
40 For presets that differ only in API keys or endpoints, use the instance's
41 **Environment variables**. Variable assignments do not belong in **Launch arguments**.
42
43 File paths in the `--settings` and `--append-system-prompt-file` launch arguments
44 can start with `~` for the home directory of the machine running T3 Code. Any
45 other relative path is read from the thread's project folder, or its worktree.
46
47 Claude Code's verbose mode can stay enabled when you use Claude for text generation, including
48 thread titles, branch names, commit messages, and pull request descriptions. On a remote connection,
49 T3 Code uses the Claude configuration on the connected server.
50
51 ## Compact long conversations
52
53 Set **Auto-compact after** in the Claude provider settings to an integer between
54 `100000` and `1000000`. For example, `300000` asks Claude to summarize at about
55 300,000 tokens. This changes when compaction happens, not the model's context
56 window. Leave it empty for Claude Code's default.
57
58 You can also send `/compact` in an existing conversation. Claude Code also
59 compacts automatically. To see context usage, see
60 [commands and skills](./composer.md#commands-and-skills).
61
62 When you return to a large thread after Claude's prompt cache expires, a
63 **Compact** chip with the thread's token count shows
64 next to the send button. While it is on, Enter summarizes the history first, then
65 sends your message. Click the chip to switch it to **Keep** and keep the full
66 history for that message. The mobile app shows no chip: an eligible message sent
67 from it summarizes first. The cache lasts about an hour on a Claude
68 subscription and 5 minutes with an API key, Bedrock, or Vertex. See [commands and skills](./composer.md#commands-and-skills) for using
69 composer commands.
70
71 **Compact idle threads** in **Settings > Providers > Advanced** controls this for threads
72 over 200,000 tokens. Each Claude provider's settings can override it:
73
74 - **When you return** shows the **Compact** chip described above. This is the default.
75 - **Before the cache expires (beta)** tries to compact an idle thread shortly before its
76 prompt cache expires, while the history is still cached.
77 The compaction appears in the thread as a `/compact` turn sent by T3 Code.
78 If the window is missed, such as when you enable this mode after the cache has
79 expired or the server restarts before compaction runs, it falls back to
80 **When you return**. Switch the chip to **Keep** to keep the full history for that message.
81 - **Never** skips idle compaction and Claude's resume prompt. **Auto-compact after**
82 still applies, and you can still compact by hand.
83
84 ## Usage limits
85
86 If your Claude subscription runs out of usage mid-turn, the thread shows which
87 limit was reached and the remaining wait when Claude provides a reset time.
88 Claude Code holds the turn until that window reopens, so it can keep showing as
89 working. Wait for the reset, or stop the turn and continue later. The warning's
90 timestamp shows when the displayed wait started.
91
92 ## Steering
93
94 A steer waits until Claude finishes the tool calls it is running, then Claude reads it
95 in the same turn. The message shows as unread until then. If Claude drops it before
96 reading it, the message is sent as the next turn, after any messages queued before it.
97 If Claude exits or the server restarts first, the message waits in the queue in that
98 same place until you resume it. If you press Stop, T3 Code shows a notice with the text
99 of any steer Claude had not read.
100
101 ## Skills
102
103 Claude skills come from the config directory's `skills` folder and the project's
104 `.claude/skills` folder. If both define the same name, the config-directory copy
105 wins. Skills disabled in Claude's settings do not appear in the composer.
106
107 Use `$` in the composer to select a skill. Enabled installed Claude plugins,
108 and plugins synced from your claude.ai account, also contribute skills under
109 names such as `plugin:review`. Skills marked `disable-model-invocation` can
110 still be started by you. Invoke those one per message: Claude directly runs only
111 the last named skill and may try to start earlier ones through its Skill tool,
112 which refuses skills reserved for manual invocation.
113
114 ## Task lists
115
116 To show Claude task lists in the web and desktop apps, add
117 `CLAUDE_CODE_ENABLE_TODO_TOOLS=1` under **Variables** in the Claude instance's
118 **Environment** section in **Settings > Providers**. Each update shows as a Tasks
119 row in the thread; open it to see every step.
120
121 ## Sign in to an MCP server
122
123 When an MCP server you added to Claude needs you to sign in, the thread shows a notice with
124 **Sign in**, once per session; on mobile, tap the notice. Approve access on the page it opens. Your
125 browser then goes to a page that may not load, especially when Claude runs on another machine:
126 copy that page's address, paste it into T3 Code, and choose **Finish sign-in**. Claude keeps the
127 sign-in and reconnects the server. Connectors you added on claude.ai are signed in on claude.ai,
128 so T3 Code doesn't ask about them. Signing in needs permission to manage providers.
129
130 ## Claude in Chrome
131
132 Claude threads can use the Claude in Chrome extension, which works in your own
133 Chrome with its sign-ins. T3 Code sessions load it only when asked: add `--chrome`
134 to **Launch arguments** in the Claude instance's settings in **Settings > Providers**.
135 It applies to sessions that start after you save. Set up the extension as you would
136 for Claude Code first; see Claude Code's [Chrome guide](https://code.claude.com/docs/en/chrome).
137
138 The instance must be signed in with `/login` on a direct Anthropic plan. Instances
139 that use an API key or a `claude setup-token` token don't get it, even with
140 `--chrome`, and neither do third-party providers such as Amazon Bedrock. Claude Code
141 doesn't support it in WSL.
142
143 ## OpenRouter
144
145 Create a Claude instance with its own config directory, such as
146 `~/.claude_openrouter`, and keep **Binary path** set to `claude`. In that instance's
147 **Environment variables**, use:
148
149 | Variable | Value |
150 | ---------------------- | ----------------------------------------- |
151 | `ANTHROPIC_BASE_URL` | `https://openrouter.ai/api` |
152 | `ANTHROPIC_AUTH_TOKEN` | Your OpenRouter API key, marked Sensitive |
153 | `ANTHROPIC_API_KEY` | An explicitly empty value |
154
155 If that Claude config directory has a cached Anthropic login, run `/logout` in a
156 Claude Code session using that directory before starting the router setup. Cached
157 login credentials can conflict with the router token.
158
159 Select the model you want in T3 Code. For an OpenRouter model outside the built-in
160 list, open that Claude instance in **Settings > Providers** and add its full model
161 ID with **Add custom model**. Then select it in the chat model picker.
162 `ANTHROPIC_DEFAULT_*_MODEL` variables map Claude Code aliases such as `sonnet`; they
163 do not replace the explicit model ID selected in T3 Code. Custom models may have
164 fewer effort, thinking, or context controls than built-in models.
165
166 Verify the model used in OpenRouter's activity dashboard. For current compatibility
167 requirements, use the
168 [OpenRouter Claude Code guide](https://openrouter.ai/docs/cookbook/coding-agents/claude-code-integration).
169
170 ## Other routers
171
172 A local router uses an ordinary Claude provider instance. Give it a separate
173 config directory and put the router's endpoint and credential variables in that
174 instance's **Environment variables**. The router must run where the environment
175 can reach it. Follow the [Claude Code Router instructions](https://github.com/musistudio/claude-code-router)
176 for its installation and routing configuration.