Go-Native AI Coding CLI by HowlerOps

Oculus

A fast, self-contained AI coding assistant that lives entirely in your terminal. Multi-model routing, native Anthropic/OpenAI/Gemini providers, CLI bridges, Charm v2 TUI, cost ceilings, fallback chains, smart model routing, and an interactive settings page — all in one binary.

Quick Install
$ go install github.com/howlerops/oculus/cmd/oculus@latest
— or —
$ brew install howlerops/tap/oculus
— or —
$ npm i -g @howlerops/oculus
$ oculus
◆ Oculus v0.6.5 ready • lens: focus • model: claude-sonnet-4-6
✓ Authenticated via OAuth
21MB
Single binary
178
Source files
44+
Slash commands
6
Provider bridges
What makes Oculus different

Built for how developers actually work

No cloud syncing, no Electron, no runtime. Just a compiled binary that talks to the AI providers you already use.

🏭

Multi-Model Lens Routing

Three purpose-built lenses — Focus for deep reasoning, Scan for fast exploration, Craft for code generation — automatically route to the right model and provider for the task.

🔌

Provider Flexibility

Anthropic, OpenAI-compatible, Bedrock OpenAI-compatible, Google AI native Gemini, Ollama (local), Claude CLI, and Codex CLI — swap providers without changing your workflow. Use native API bridges where possible and subscription-backed CLI bridges where they offer the strongest parity.

🔨

40 Built-In Tools

Bash execution, file read/write/edit/glob/grep, web search, sub-agent dispatch, git operations, LSP integration, and more — all wired directly into the model's tool-use protocol.

📌

Episode-Based Context

Conversations are persisted as typed episodes. LCM dual-threshold compaction automatically summarises old context, keeping the window fresh without losing history.

💻

Full Terminal UI

Charm v2 TUI with markdown rendering, syntax-highlighted code blocks, interactive settings page, permission dialogs, model picker, vim keybindings, and a two-section status bar.

🔒

Safeguards & Security

Cost ceilings, step limits, query duration caps, permission pre-flight, fallback chains, and boulder recovery. OAuth PKCE with OS keychain. Bash safety analysis with 24 dangerous patterns.

Zero to first response in 60 seconds

Up and running fast

Three install paths, one command to start. Oculus auto-detects your API keys and walks you through auth on first run.

😺
Go go install ...@latest
🍺
Homebrew brew install ...
📦
npm npm i -g ...
oculus — interactive session
$ oculus
◆ Oculus v0.6.5 lens:focus model:claude-sonnet-4-6 tools:40

> Refactor the auth package to use interface injection
◆ Reading pkg/auth/...
◆ Analysing 6 files, 842 lines
✓ Here's the refactored interface — 3 files changed, tests updated.

> oculus -p "explain the lens routing system" --lens scan
◆ Scan lens active (fast exploration mode)
Full Getting Started Guide →
Provider support

Your models. Your infrastructure.

Oculus ships bridges for every major AI provider and supports fully local inference through Ollama.

API

Anthropic

Claude Opus 4.6, Sonnet 4.6, Haiku 4.5. Native streaming, extended thinking, and tool-use.

API

OpenAI-Compatible

GPT-4o, o3, and any provider exposing the OpenAI Chat Completions spec.

Local

Ollama

Run Llama 3, Mistral, Qwen, and hundreds of open models entirely on-device.

CLI

Claude CLI

Delegates to the Anthropic claude CLI binary when present.

CLI

Codex CLI

OpenAI Codex bridge via subprocess, with streamed output forwarding.

CLI

Google AI / Gemini

Native Gemini API bridge via the official Go SDK. Gemini CLI remains available as a lighter-weight bridge when you want to use the local CLI directly.

Ready to try Oculus?

Install in seconds. No sign-up required — just bring your own API key or use an Ollama local model.

Get Started → View on GitHub Architecture Docs