Argus
Download

AI workspace for Unity and game teams

Argus gives game projects a working memory.

Build with a local Project Brain, approval-gated tool execution, Unity context, MCP support, and provider controls that separate private local runs from higher-quality cloud model work.

Argus
Project Brain Indexed
Developer

Use Unity context to patch the failing inventory serializer.

Argus

Hydrating scene diagnostics, code symbols, graph context, and recent changes.

Ask, plan, or run an agent task...
01 Project Brain that learns the work

Indexes code, docs, decisions, snippets, graph context, and recent file changes across sessions.

02 Unity-aware agent workflow

Connects editor status, scene and selection diagnostics, command results, and support evidence.

03 Local privacy, cloud quality when needed

Run locally with Ollama or LM Studio, then escalate hard tasks through provider CLI or API lanes.

Feature surface

Built for project-aware game development, not blank chat.

Argus keeps the model connected to the project, the engine, and the approval path so agent work stays inspectable before anything changes.

Project Brain

Builds durable project memory from code, docs, decisions, snippets, diagnostics, recent changes, and approved team knowledge.

Unity context bridge

The Unity package connects editor status, command queues, scene and selection diagnostics, asset context, and result posting.

Argus Tool Gateway

File, terminal, and engine-facing work runs through permissions, signed command envelopes, policy checks, and visible proof.

Graph and RAG context

Semantic chunks, local vector and text search, graph neighborhoods, hierarchy summaries, and context packs help models retrieve the right code.

Provider choice

Use Ollama as the production local path, optional LM Studio for local OpenAI-compatible serving, or provider CLI/API lanes for stronger models.

MCP operating layer

Advanced MCP server support exposes tools, resources, prompts, traces, OAuth metadata, and client profiles without advertising unreleased plugins.

Project Brain

Context that grows from the actual project.

Argus prepares context locally first. Provider-visible context is selected through policy and approval instead of pushing raw project state by default.

Incremental project learning

The indexer tracks file fingerprints, changed paths, queue state, code symbols, docs, and deferred changes so later tasks start from current context.

RAG with source lineage

Code and Markdown are split into semantic chunks with file, symbol, section, and line metadata, then searched through local vector and FTS stores.

Graph-guided context packs

Concept graphs, neighborhood lookahead, hierarchical summaries, and hot cache hydration assemble bounded context for the task at hand.

Memory lifecycle

Approved and proposed memories can cover project facts, rules, decisions, snippets, diagnostics, episodic events, and visual ingest results.

Promotion and contradiction checks

Useful memories can be promoted across conversation, session, project, workspace, and user scopes while contradiction checks keep stale facts visible.

Provider-ready handoff

The brain builds audited, token-bounded handoff payloads for local or cloud providers after context visibility rules are applied.

Workflow

From project context to reviewable engine work.

The app works like a command surface for serious changes: hydrate the project, choose the provider, run through the gateway, and keep proof attached.

argus/workspace context

$ project brain hydrate --unity

Indexed fingerprints, semantic chunks, graph neighborhoods, and recent changes.

Context pack prepared locally; provider-visible context waits for approval.

Model layer

Choose local privacy or cloud quality with the tradeoff visible.

Default local Ollama for private local runs and headless local workflows
Optional local LM Studio through a local OpenAI-compatible server
Cloud CLI Codex CLI/App Server and Gemini CLI for stronger agent work
API credits OpenAI API support for advanced or admin-managed usage
Reality check Local models are useful, but top cloud models handle hard engine tasks better
OpenArgus

Provider login stays on official paths.

Argus supports provider-owned CLI login and API-credit lanes through official auth paths. Use local models when privacy matters; use higher quality cloud providers when the task needs deeper reasoning.

MCP

An MCP layer for controlled developer tools.

Argus exposes MCP server surfaces and client profiles for tool-aware workflows while keeping mutating actions behind ask-first policy.

Server and transport support

Streamable HTTP, SSE-style endpoints, sessions, traces, and protected resource metadata are part of the MCP surface.

Tools, resources, and prompts

MCP clients can discover Argus tools, resources, prompts, completion, progress, and cancellation support through defined profiles.

No plugin overclaiming

The website should describe the MCP operating layer and verified client paths, not unreleased external plugins or embedded third-party agents.

Unity integration

Editor context for engine-side work.

Package Unity Editor bridge with managed DLL and native cores
Context Scene, selection, asset diagnostics, editor status, and command results
Mutation safety Non-dry-run commands require Argus-signed command envelopes
Support bundle Diagnostics, queue state, serializer coverage, package version, and redacted auth status
Unity-first game workflows

Bring project learning into the editor loop.

Argus can reason over code and engine context together, then route file and editor-facing work through approvals, command verification, and visible diagnostics before changes are trusted.

Downloads

Install Argus where the work happens.

Windows, macOS, and Linux installers ship together as one desktop release set. Unity integration extends Argus into engine-side work.

Platforms

macOS

Apple Silicon and Intel builds for local-first workflows.

Download macOS
Game engines

Unity

Connect scenes, scripts, editor actions, and project context.

Managed through Unity