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

main
md 178 lines 8.19 KB
Rendered Raw
1 # Codex
2
3 Use your ChatGPT plan or an existing Codex CLI login to code in T3 Code.
4
5 ## Connect with ChatGPT
6
7 Connect during onboarding or in **Settings → Providers**. For a remote machine,
8 select that environment first. T3 Code handles Codex installation; sign in on
9 OpenAI and allow sharing of your ChatGPT plan.
10
11 Manage shared usage and credits in ChatGPT through **Manage usage** in T3 Code.
12 If a request uses a feature that ChatGPT sharing does not support, use another
13 provider for that request.
14
15 Fast and Ultrafast speeds are unavailable with this connection. To use the speed
16 options your account supports, [use an existing Codex login](#use-an-existing-codex-login).
17
18 When reconnecting, choose the same account in T3 Code and on OpenAI's sign-in
19 page. Disconnecting stops running threads but keeps their history and lets you
20 reconnect later.
21
22 If remote sign-in cannot return automatically, paste the full URL from the final
23 localhost page into the sign-in panel, even if that page could not load.
24
25 ## Use an existing Codex login
26
27 T3 Code can use your installed Codex and its existing login. Run `codex login`
28 on the environment's machine to sign in. [Provider setup](./install.md#providers)
29 covers installation and custom configuration.
30
31 Running `codex login` does not change a managed ChatGPT connection. In
32 **Settings > Providers**, add a Codex instance, choose **Configure manually**,
33 and complete the wizard. Then select that instance from the thread's model
34 picker. See [Use multiple accounts](#use-multiple-accounts).
35
36 ## Allow more time for provider checks
37
38 If an authenticated Codex installation reports “Timed out while checking Codex
39 app-server provider status,” set `T3CODE_CODEX_PROBE_TIMEOUT_MS=60000` in the
40 T3 server's environment and restart it. See [Provider check timeouts](./install.md#provider-check-timeouts)
41 for configuration and other providers.
42
43 ## Generate images
44
45 Codex can generate images directly in a T3 Code thread when your Codex version,
46 model, and account support native image generation.
47
48 If Codex reports that its image tool is unavailable, image generation may be
49 disabled in its configuration. In **Settings → Providers**, edit your existing
50 Codex provider and add this to **Launch arguments**, keeping any other arguments:
51
52 ```text
53 -c features.image_generation=true
54 ```
55
56 This enables the feature for that T3 Code provider without changing your shared
57 Codex configuration. You do not need another provider instance or a separate image
58 API key. The flag does not add image support to an unsupported model or account.
59
60 Launch arguments apply to Codex sessions started after you save them. Start a new
61 thread with the same Codex provider and ask it to generate an image. The completed
62 image appears in the thread.
63
64 ## Use multiple accounts
65
66 Add another ChatGPT account in **Settings → Providers**, then select the account
67 from the thread's model picker. Compatible accounts can continue the same thread.
68 Connecting accounts through T3 Code leaves your CLI login unchanged.
69
70 ### Multiple CLI logins
71
72 A shared Codex home with a shadow home lets work and personal accounts continue
73 the same threads. The accounts share Codex sessions and configuration while keeping
74 their own login and available models.
75
76 Keep your first account in `~/.codex`. On the environment's machine, sign the
77 second account into a fresh directory:
78
79 ```bash
80 mkdir -p ~/.codex_personal
81 CODEX_HOME=~/.codex_personal codex login
82 ```
83
84 Then add a second Codex instance in **Settings > Providers**:
85
86 | Instance | CODEX_HOME path | Shadow home path |
87 | -------------- | --------------- | ------------------- |
88 | Codex Work | `~/.codex` | Leave empty |
89 | Codex Personal | `~/.codex` | `~/.codex_personal` |
90
91 Both instances must use the same **CODEX_HOME path**. T3 Code prepares the shared
92 state in the shadow directory; do not populate it by copying your whole Codex
93 home.
94
95 The shadow account needs its own `auth.json` file. If Codex uses an OS credential
96 store, configure file storage for this setup. See
97 [OpenAI's credential storage guide](https://learn.chatgpt.com/docs/auth#credential-storage).
98
99 Use a completely separate **CODEX_HOME path**, with no shadow home, when you want
100 separate Codex sessions and configuration. That instance cannot continue threads
101 from the other home.
102
103 ## Switch accounts in an existing thread
104
105 Choose the other account from the thread's model picker. T3 Code offers compatible
106 Codex instances that share the thread's **CODEX_HOME path**. Changing accounts does
107 not move the conversation into a separate Codex home.
108
109 If the account is missing from the picker, compare the home paths in provider
110 settings. If two instances show the same unexpected account or models, check their
111 reported accounts, refresh provider status, and confirm the second instance has
112 its own shadow path and login. A shadow-home conflict usually means the directory
113 contains a copied Codex setup. Use a fresh shadow directory and sign in again.
114
115 ## Answer questions while Codex works
116
117 Codex can ask a question and keep working. Answer it in the thread's question
118 panel. The answer becomes a new message: it reaches the active turn, or starts
119 another turn if Codex has finished. Unanswered questions survive reconnects.
120 If you do not want to answer, dismiss the question from its panel. Dismissing
121 closes it without sending anything to Codex. This requires a Codex version that
122 supports async questions.
123
124 ## Approve app access
125
126 Codex tools can request access to another app. Respond to the named app's request
127 in the thread on web, desktop, or mobile. Some tools offer access for one request,
128 the current session, or permanently. See [Permission modes](./permission-modes.md)
129 for command and file approvals.
130
131 ## Sign in to an MCP server
132
133 When an MCP server you added to Codex needs you to sign in, the thread shows a notice with
134 **Sign in**; on mobile, tap the notice. Approve access on the page it opens. Your browser then goes
135 to a page that may not load, especially when Codex runs on another machine: copy that page's
136 address, paste it into T3 Code, and choose **Finish sign-in**. Codex reconnects the server in that
137 thread and keeps the sign-in for other threads on the same Codex instance; an instance with its own
138 home, such as a second account, may need to sign in separately. Signing in needs permission to
139 manage providers.
140
141 ## Use the OpenAI Chrome extension
142
143 Connecting ChatGPT in **Settings → Providers** signs in the Codex provider.
144 Using your signed-in Chrome profile through the extension requires a browser
145 connection for the current Codex thread.
146
147 If a T3 Code thread reports `The browser access host is unavailable`, its browser
148 host is unavailable; this does not identify an extension connection failure.
149 For a task that requested Chrome, follow [OpenAI's setup guide](https://learn.chatgpt.com/docs/chrome-extension),
150 then select Chrome in a Codex chat in the ChatGPT desktop app.
151 If the task runs on a remote environment, `localhost` in the desktop browser
152 refers to your computer. Forward the remote page's port or use a URL reachable
153 from your computer before repeating the task. Connecting to T3 Code through a
154 relay or tunnel does not by itself expose that page. T3 Code's built-in browser
155 preview uses its own browser session.
156
157 ## Use Daybreak
158
159 Models with cyber access programs show a Cyber Access option in the model
160 options. It starts on Standard, or on Daybreak when `config.toml` in your Codex
161 home sets `daybreak = true` or the model only runs under Daybreak. Your choice carries
162 over to new threads. Custom models do not offer it. If Codex rejects a turn
163 because your login lacks the program, switch Cyber Access back to Standard or
164 sign in to Codex with an account that has it.
165
166 ## Codex says I hit a usage limit
167
168 When Codex stops on a usage limit, the thread names the window that ran out and
169 when it resets, when Codex reports them. Send the message again after the reset. On a workspace plan the
170 message also says whether your workspace owner needs to add credits or raise the
171 spend limit to continue sooner.
172
173 ## Send feedback to OpenAI
174
175 In an existing Codex thread, send `/feedback` with an optional description, for
176 example `/feedback The agent stopped before finishing the tests`. This uploads
177 the conversation and Codex logs to OpenAI. The returned thread ID can be shared
178 with OpenAI support.