echoVic/orca-agent

Orca is a DeepSeek-native coding agent.

What is Orca?

Orca is a locally‑run, terminal‑based coding assistant that talks to the DeepSeek large‑language model. It can read, edit, and create source files, run shell commands, and verify the results – all from a text‑based UI or in a headless “exec” mode that works in CI pipelines.

How you use it

  • Install – via an npm global package (npm i -g @blade‑ai/orca) or by downloading a pre‑built Rust binary.
  • Run – orca opens an interactive TUI; orca exec "fix the failing test" runs a one‑off task without UI. Sessions can be saved, resumed, or forked.
  • Give it a task – type a natural‑language instruction. Orca turns the instruction into a plan, calls tools (file edit, shell commands, image handling, etc.), shows the model’s reasoning, and asks for your approval when needed.
  • Control – shortcuts let you approve tool calls, interrupt, background work, view sub‑agents, and manage “goals” that persist across turns.

Key features (from the README)

Feature What it means
DeepSeek‑native Uses DeepSeek‑Flash or DeepSeek‑V4‑Pro models with streaming output and optional visual‑analysis of images.
Terminal UI Rich TUI with mentions (@), skill insertion ($), command menu (/), and image paste (Ctrl+V).
Headless / CI mode orca exec can be scripted; it prints a resume command so a later run can continue the same session.
Tool sandboxing Actions are gated by modes – suggest, auto‑edit, full‑auto, plan – and per‑folder trust settings.
Sub‑agents & workflows Spawn synchronous or asynchronous child agents, each with its own conversation and controls.
Persistent memory Learns project‑specific facts after successful turns and can recall them later.
MCP / ACP integration Connects to external “MCP” servers (e.g., Pilion Browser) and exposes an Agent Client Protocol for editors or other harnesses.
Cross‑platform Binaries for macOS, Linux, Windows (ARM & x64); also distributed via npm.
Configurable Settings via environment, CLI flags, or ~/.orca/config.toml.
Reliability Handles cancellation, checkpointing, and continuation recovery; isolates long‑running commands from the UI.

Who might find Orca useful?

  • Developers who want an AI assistant that can modify code on their own machine without sending source to the cloud.
  • CI/CD pipelines that need an automated “fix‑it” step – orca exec can run in a container and resume if it hits a budget limit.
  • Power‑users who like a terminal‑first workflow but still want the model to see images or interact with a browser (via Pilion).

Where to learn more


All information above is taken directly from the project's README; no additional claims are made.

Related

  • Project
  • Project
  • Project
  • Project
  • Project