DevIntern does not ship its own model or agent. It formats tracker context, invokes the coding agent you already use, and handles branches, self-review, and pull requests around each run. Swapping agents is one line: set AGENT_HARNESS in your project configuration and everything else stays the same.
Each agent below works with every supported task tracker, in both @devintern/code and @devintern/pm. Runs execute on your machines with your own model keys; DevIntern does not host or proxy model calls.
OpenCode
AGENT_HARNESS=opencode
Use OpenCode as your coding agent for both ticket drafting and implementation. DevIntern handles prompt formatting, branch management, and git artifacts around each OpenCode run.
Setup: Install OpenCode and set AGENT_HARNESS=opencode.
Drop-in alternative to Claude Code via one env var
Same tracker and repo integrations as other harnesses
Headless runs suitable for server automation
Bring your own OpenCode model configuration
How do I switch from Claude Code to OpenCode?
Set AGENT_HARNESS=opencode in .devintern-code/.env and .devintern-pm/.env. No other workflow changes are required.
Grok Build
AGENT_HARNESS=grok
Run xAI Grok Build as the agent behind planning and implementation. DevIntern invokes the grok CLI headlessly with -p so ticket runs complete without an interactive TUI.
Setup: Install Grok Build (curl -fsSL https://x.ai/cli/install.sh | bash) and set AGENT_HARNESS=grok.
Headless Grok Build via -p (single-prompt mode)
Optional --always-approve for unattended tool execution
Model override via -m / harness configuration
Works with every supported task tracker
Is the CLI named grok or grok-build?
The product is Grok Build; the executable on PATH is grok. Set AGENT_HARNESS=grok (not grok-build).
Does --max-turns work with Grok?
No. Grok Build does not expose a max-turns flag. DevIntern still runs the task to completion under the harness timeout.
Reasonix
AGENT_HARNESS=deepseek
Use Reasonix (DeepSeek-Reasonix) as your DeepSeek-native coding agent for both ticket drafting and implementation. DevIntern handles prompt formatting, branch management, and git artifacts around each reasonix run.
Setup: Install Reasonix (npm i -g reasonix) and set AGENT_HARNESS=deepseek.
Harness id deepseek maps to the reasonix binary
Headless reasonix run suitable for automation
Model override via --model
Listed in DeepSeek's official agent integrations docs
Is Reasonix an official DeepSeek product?
Reasonix is a community DeepSeek-native coding agent listed in DeepSeek's agent integrations docs. There is no official DeepSeek-branded coding CLI yet. Set AGENT_HARNESS=deepseek to use it.
Do skip-permissions or max-turns flags apply?
reasonix run is already autonomous for writer tools (deny rules still apply). Turn limits are configured in Reasonix config (max_steps), not via DevIntern --max-turns.
Antigravity CLI
AGENT_HARNESS=antigravity
Run Google's Antigravity CLI (agy) as the agent behind planning and implementation. DevIntern invokes agy -p headlessly so ticket runs complete without an interactive session.
Setup: Install Antigravity CLI (curl -fsSL https://antigravity.google/cli/install.sh | bash on macOS/Linux) and set AGENT_HARNESS=antigravity.
Headless Antigravity CLI via agy -p (single-prompt mode)
Optional --dangerously-skip-permissions for unattended tool execution
Works with every supported task tracker
Same self-hosted security model as other harnesses
Does --max-turns work with Antigravity CLI?
No. Antigravity CLI does not expose a max-turns flag. DevIntern still runs the task to completion under the harness timeout (AGENT_HARNESS_TIMEOUT_MINUTES).
I still have AGENT_HARNESS=gemini. Does it work?
Yes for now. DevIntern maps gemini to the Antigravity harness and prints a deprecation warning. Install agy, set AGENT_HARNESS=antigravity (or agy), and re-auth. Consumer Gemini CLI stopped serving on 2026-06-18; DevIntern does not spawn the gemini binary.
How do I authenticate for headless or CI runs?
Locally, agy uses browser OAuth and stores credentials in the OS keyring. For non-interactive environments, use the Antigravity token flow documented by Google (ANTIGRAVITY_TOKEN). Legacy GEMINI_API_KEY is ignored.
Cline
AGENT_HARNESS=cline
Use Cline as your coding agent for both ticket drafting and implementation. DevIntern handles prompt formatting, branch management, and git artifacts around each cline task run.
Setup: Install the Cline CLI (docs.cline.bot) and set AGENT_HARNESS=cline.
Headless cline task runs with your tracker context
Optional --yolo for unattended tool execution
Model override via --model / harness configuration
Works with every supported task tracker
How do I switch to Cline?
Set AGENT_HARNESS=cline in .devintern-code/.env and .devintern-pm/.env. No other workflow changes are required.
Goose
AGENT_HARNESS=goose
Run Block's Goose as the agent behind planning and implementation. DevIntern invokes goose run headlessly; Goose's autonomous Auto Mode fits unattended ticket runs out of the box.
Setup: Install Goose and configure your provider and model.
Headless goose run -t execution
Autonomous Auto Mode by default, no permission prompts to script around
Bring your own provider and model via Goose configuration
Works with every supported task tracker
Which models does Goose use?
Whatever provider and model you configure in Goose itself. DevIntern passes the task prompt and leaves model routing to your Goose configuration.
Kilo Code
AGENT_HARNESS=kilo-code
Point DevIntern at Kilo Code for implementation and spec drafting. One AGENT_HARNESS=kilo-code value swaps the agent while your tracker, repo, and review workflow stay the same.
Setup: Install the Kilo Code CLI and set AGENT_HARNESS=kilo-code.
Harness id kilo-code maps to the kilo binary (like cursor maps to cursor-agent)
Headless kilo run with optional --auto for unattended execution
Model override via -m / harness configuration
Works with every supported task tracker
Why is the harness called kilo-code but the binary is kilo?
kilo-code is the product-facing harness id. The installable CLI is kilo. Set AGENT_HARNESS=kilo-code; optionally set AGENT_CLI_PATH if kilo is not on PATH.
Kimi CLI
AGENT_HARNESS=kimi
Run Moonshot AI's Kimi CLI as the agent behind planning and implementation. DevIntern invokes kimi --print headlessly so ticket runs complete without an interactive session.
Setup: Install the Kimi CLI and authenticate with your Moonshot AI account.
Headless kimi --print execution
Optional --yolo for unattended tool execution
Model override via --model / harness configuration
Works with every supported task tracker
How do I switch to Kimi CLI?
Set AGENT_HARNESS=kimi in .devintern-code/.env and .devintern-pm/.env. No other workflow changes are required.
Qwen Code
AGENT_HARNESS=qwen
Use Qwen Code as your coding agent for both ticket drafting and implementation. DevIntern invokes qwen in headless mode and handles branch management and git artifacts around each run.
Setup: Install Qwen Code (npm install -g @qwen-code/qwen-code) and set AGENT_HARNESS=qwen.
Headless Qwen Code via -p (single-prompt mode)
Optional --yolo for unattended tool execution
Model selection via Qwen Code settings (config-file based)
Works with every supported task tracker
Can DevIntern override the Qwen model per run?
No. Qwen Code does not expose a --model CLI flag today. Configure your model in ~/.qwen/settings.json and DevIntern uses whatever Qwen Code is set to.
Get started
Pick an agent and run your first ticket
Install your preferred agent, point @devintern/code at a ticket, and review the pull request it opens.