Skip to main content
DeepCLI is the DeepSeek-oriented command-line interface for the Termux Monorepo. It lets you create conversations, send prompts with optional reasoning mode, list and browse session history, fork threads at any message, and export chats in multiple formats. DeepCLI also integrates with GitHub Actions to trigger AI workflows on pull request events.

​
Entry points

DeepCLI exposes four primary entry points in the deepcli/ directory:
  • deepcli.py — Main interactive CLI for session management
  • deepapi.py — Programmatic API wrapper for scripted access
  • deepseek_proxy.py — Proxy layer for request routing and rate handling
  • ci_mode.py — CI/CD integration script for GitHub Actions

​
Session commands

Use the following commands to create, interact with, and manage DeepSeek sessions.

​
Thinking mode

Add the --thinking flag to any send command to enable DeepSeek reasoning mode. This asks the model to expose its chain of thought before delivering the final answer, which is useful for debugging prompts or validating complex reasoning tasks.

​
Conversation forking

Forking lets you branch a conversation at an earlier message without losing the original thread. Specify the session ID and the message ID where you want the branch to start. The forked session receives a new ID and inherits all messages up to the fork point.

​
Export formats

DeepCLI supports exporting session history for archival or further processing. Use the export command with --format json to write a structured JSON file, or specify another supported format. The --output flag controls the destination path.

​
CI/CD integration

DeepCLI integrates with GitHub Actions through ci_mode.py. Configure your workflow to trigger on pull request events (open, synchronize, close). The CI script can automatically send prompts, export results, and post summaries back to the pull request thread.
1

Add the workflow file

Create .github/workflows/deepcli.yml in your repository and reference deepcli/ci_mode.py.
2

Set the trigger events

Configure the workflow to run on pull_request events: opened, synchronize, and closed.
3

Pass secrets

Provide the DEEPSEEK_API_KEY as a repository secret so ci_mode.py can authenticate.
4

Review outputs

The workflow exports session results as artifacts and optionally posts a summary comment to the PR.
CI mode uses deepseek_proxy.py for request routing, which helps manage rate limits and retries during automated runs.