Orca
Run BlazeInfer-powered coding agents side by side, each in its own git worktree.
Orca is an open-source desktop app that runs many CLI coding agents side by side, each in its own git worktree, with diff review and issue-tracker integration.
Orca does not call a model API itself
Orca launches agent processes in terminals. To use BlazeInfer, you configure the agent (OpenCode, Kilo Code or Pi) for BlazeInfer, then let Orca run that agent.
Prerequisites
- A BlazeInfer account with credits and an API key.
- An agent that speaks OpenAI Chat Completions and is configured for BlazeInfer. See OpenCode, Kilo Code or Pi. Check that the agent works with BlazeInfer in a normal terminal before you add it to Orca.
Install
On macOS, install Orca with Homebrew:
brew install --cask stablyai/orca/orcaUse the full name. brew install --cask orca installs a different, unrelated
app. You can also download Orca for macOS or Linux from its
GitHub releases.
Configure
- Set up the BlazeInfer provider in the agent's own config file (for example
~/.config/opencode/opencode.json). - Open Orca. On first start it asks you to pick a default agent. Choose the one
you configured for BlazeInfer. OpenCode 2.x installs an
opencode2command, which Orca lists as OpenCode 2; pick the entry whose command matches the version you set up. - In Settings → Agents, expand the agent's row. It shows the agent's
Command, Permissions, Arguments and Environment.
- If the agent's config reads the key from the environment (for example
"apiKey": "{env:BLAZE_API_KEY}"), addBLAZE_API_KEYwith yourblz_...key to Environment, or rely on the login-shell environment that agents inherit. - If the key is written in the agent's config file, you can leave Environment empty.
- If the agent's config reads the key from the environment (for example
- Restart any open sessions of that agent.
Yolo mode
Orca's first-start screen turns on Yolo mode, and the default Permissions is Full access. Agents then run commands and edit files without asking. Use this only in projects you trust.
Verify it works
- Add a project in Orca (any folder that is a git repository), then create a
worktree with your BlazeInfer-configured agent. Orca creates it under
~/orca/workspaces/<project>/<workspace>. - In the agent, choose a BlazeInfer model, for example
deepseek-ai/deepseek-v4.1-flashorzai-org/glm-5.3-flash(in OpenCode, type/models). - Send:
create hello.py that prints hi, then run it.
A working setup creates hello.py, runs it and prints hi. The new file also
appears in Orca's file list.
Configuration tips
- Restart after changes. Agent sessions must be restarted after changing config or environment.
- Permission mode. A custom argument or environment override takes that agent out of Orca's global permission-mode switching.
- Remote worktrees. On SSH worktrees or a remote Orca server, the agent
config and
BLAZE_API_KEYmust exist on that host, not just your laptop.