# Native chat: talk to your coding agents, drop to the terminal anytime

> Native chat in Tallos is a chat view layered on top of a coding agent's terminal session. Supported agents such as Claude Code, Codex and Grok CLI open in chat by default, with messages, tool calls, file diffs and questions laid out for reading. The terminal stays the source of truth: View terminal shows the raw session, Back to chat returns.

- Canonical: https://runtallos.com/features/native-chat
- Português: https://runtallos.com/pt/recursos/chat-nativo.md
- Updated: 2026-09-29
- Section: Features

## Key facts

- Default view for Claude Code, Codex, Grok CLI, OpenClaude and OMP
- View terminal / Back to chat on every agent pane
- Settings → Agents → Agent View: Chat or Terminal
- Question and approval cards you answer in place
- Model and effort pills, slash commands and skills

## What is native chat in Tallos?

**Native chat** is a structured transcript and composer that sits over the same terminal session an agent CLI runs in. The agent is still the vendor's own CLI, running in a real terminal, signed in with your own account. Tallos reads the agent's transcript and renders it as a conversation, so you read messages and results instead of scrolling raw terminal output.

Because chat and terminal share one session, nothing is lost when you switch. Type in chat and the message goes to the CLI; switch to the [terminal](https://runtallos.com/features/terminal) and you see exactly where the agent is. If you are new to agent CLIs, start with [what AI coding agents are](https://runtallos.com/what-are-ai-coding-agents).

## Which agents open in chat?

Chat is the default view for new sessions of agents whose transcripts Tallos can read. Every other agent, and plain shell tabs, keep opening in the terminal.

| Agent | Opens in | Notes |
| --- | --- | --- |
| [Claude Code](https://runtallos.com/integrations/claude-code) | Chat | Model list comes from the Claude CLI installed on that machine |
| [Codex](https://runtallos.com/integrations/codex) | Chat | Pick a model straight from the model pill |
| [Grok CLI](https://runtallos.com/integrations/grok-cli) | Chat | Model and reasoning effort, with models listed by the signed-in CLI |
| OpenClaude | Chat | Uses the same transcript format as Claude Code |
| OMP | Chat | Opens in chat like the others |
| Gemini CLI, OpenCode and other agent CLIs | Terminal | Full terminal session, same workspace features |
| Plain shell tabs | Terminal | No agent, no chat |

_Where new agent sessions open in Tallos_

## How do I switch between chat and the terminal?

Use the switch in the agent pane header. **View terminal** shows the raw terminal UI of that session; **Back to chat** returns to the transcript. The switch is per pane, so you can keep one agent in chat and another in the terminal.

1. **Per session** — Click **View terminal** or **Back to chat** in the pane header. The agent keeps running either way.
2. **For every new session** — Open **Settings → Agents → Agent View** and choose **Chat** or **Terminal**. Terminal changes only where new tabs land; the per-pane switch stays available.
3. **Advanced options** — Extra chat options, such as resuming after a restart, live under **Settings → Experimental → Chat UI**.

> **When to use the terminal** — Chat is still being tuned for full terminal parity. When you need every raw status detail an agent prints, switch to the terminal for that moment and come back.

## What does a chat session show?

A new chat opens with one question: **What should Claude Code work on?** (with your agent's name). Describe what you want in plain words and the agent reads your project and makes the changes. From there, the transcript lays out each turn:

| In the transcript | What it shows |
| --- | --- |
| **Messages** | Your prompts and the agent's replies, with copy buttons for messages and code blocks |
| **Tool calls** | Grouped summaries such as "Read 4 files" or "Ran 3 commands and used 2 tools", expandable to each call and its result |
| **File edits** | Edited, added, deleted or renamed files with their diffs, plus a per-turn count of changed files |
| **Task list** | The agent's plan with pending, in-progress and completed tasks |
| **Subagents and background tasks** | Agents and commands the session started, with their state and a Stop control |
| **Context usage** | How much of the context window the session has used, and a notice when context is compacted |
| **Turn status** | Working, thinking, and how long each turn took |


> Download Tallos and open your first agent in chat. → https://runtallos.com/signup

## How do I answer an agent's questions in chat?

When an agent asks a structured question, chat renders it as a card in the transcript instead of a collapsed tool call. Pick an option, choose **Other…** to type your own answer, and confirm with **Submit**. Multi-step questions add **Next** and **Skip**.

Permission requests work the same way: an approval card asks **Allow …?** with **Allow** and **Deny**. Once answered, the card shows a receipt of what was decided, so the transcript stays a clear record of the session.

## What can I do from the chat composer?

- **Send or stop**: send a message, or stop a turn that is running.
- **Attach files and images**: attach, paste or drop files; dropped files are added to your message as paths the agent can read.
- **Slash commands and skills**: type `/` for the agent's commands and your installed skills, grouped by project, personal, built-in and plugin.
- **Model and effort pills**: change the model, reasoning effort and other session options when the agent exposes them. Some apply when the session starts.
- **Drafts survive reconnects**: unsent text and attachments stay with the pane.

## Can I resume a past agent session?

Yes. Open the right sidebar, switch to the **Agents** tab and use **Agent Session History**. It lists past sessions that agent CLIs saved on this machine, searchable by title, folder, branch, model or text. Pick one and click **Resume**: Tallos opens a new terminal in the session's folder and runs the agent's own resume command, such as `claude --resume <id>` or `codex resume <id>`. Supported agents then show that session in chat.

Every session still runs inside its own [parallel workspace](https://runtallos.com/features/parallel-workspaces), so the changes an agent makes in chat show up in that workspace's diff, ready for [diff review](https://runtallos.com/features/diff-review).

## Frequently asked questions

### Does native chat replace the Claude Code or Codex CLI?

No. Tallos runs the vendor's own CLI in a real terminal and draws a chat view over that same session. You can switch to the raw terminal at any time with View terminal.

### Which agents support chat view in Tallos?

Claude Code, Codex, Grok CLI, OpenClaude and OMP open in chat by default. Other agents, such as Gemini CLI and OpenCode, open in the terminal.

### How do I make agents always open in the terminal?

Go to Settings → Agents → Agent View and choose Terminal. New agent tabs then open in the terminal, and you can still switch any chat-capable session with Back to chat.

### Can I answer an agent's questions without using the terminal?

Yes. Questions appear as cards in the transcript: pick an option or type your own answer and click Submit. Permission requests appear as Allow or Deny cards.

### Can I change the model from the chat?

Yes, when the agent exposes it. Model and effort pills sit in the composer; for Claude Code the list comes from the installed CLI, and Grok CLI lists its models from the signed-in CLI.

### Is anything lost when I switch from chat to the terminal?

No. Chat and terminal are two views of the same session, so the agent keeps running and its full history stays in the terminal.

---

**Run your agents in parallel with Tallos** — Claude Code, Codex, Gemini and 30+ agents — each in its own workspace, with squads, chat, terminals and review in one app. Uses the subscriptions you already have.

https://runtallos.com/signup (macOS 13+ · Windows 10+)