Argentum v0.0.9

Argentum v0.0.9

Argentum v0.0.9 is a published release from the development branch. It contains the
desktop shell, provider setup, local model controls, security hardening, migration tools,
and release automation described below.

Release assets

The release workflows produce the following assets when their builds complete:

• Windows: Argentum_0.0.9_x64-setup.exe and Argentum_0.0.9_x64_en-US.msi
• macOS: Argentum_0.0.9_x64.dmg and Argentum_0.0.9_aarch64.dmg
• Linux: Argentum_0.0.9_amd64.AppImage, Argentum_0.0.9_amd64.deb, and
  Argentum_0.0.9_x86_64.rpm
• Portable CLI: argentum-v0.0.9-linux-x64, argentum-cli-v0.0.9-win-x64.exe, and
  argentum-v0.0.9-macos-x64
• Android: argentum-0.0.9-android.apk

The published files are attached to the v0.0.9 GitHub release.

Included functionality

Area	Release behavior	Status for v0.0.9
Desktop assets	Skills catalog and localization modules are generated into the Tauri frontend and checked for drift.	Implemented and covered by module-resolution tests
Startup diagnostics	The desktop shell catches renderer import, runtime, and unhandled promise failures and presents actionable details instead of leaving a blank view.	Implemented in the desktop shell and native runner
Localization	English/Estonian catalogs, locale persistence, formatters, and text-direction metadata are packaged.	Foundation only; translating every visible desktop string is planned
Thinking levels	Claude 4 receives manual extended-thinking budgets; Codex and OpenAI receive provider-specific effort fields.	Implemented; Anthropic remains a testing provider
Built-in skills	The catalog lists browser automation, computer control, skill loader, and YouTube Shorts guidance—the SKILL.md files present in this repository.	Implemented; feature dependencies and enablement still apply
OpenClaw migration	Only allowlisted item IDs are accepted. Skills merge without overwriting conflicts; memory, persona, config, credentials, database, and other files are archived for review.	Implemented; archived settings and memories are not activated automatically
Update check	GitHub Releases is checked with numeric semantic-version ordering; the action opens the releases page.	Implemented; no unattended installer
Router Agent	YAML and nested JSON rules are validated and the selected route is logged before the existing agent handles the message.	Decision/telemetry layer only
Knowledge Graph	GraphBackend is exported through a shared type module.	Consolidation groundwork only
Feedback and help	GitHub links open the tracked development docs; issue templates include version and workspace presence without exposing its path.	Implemented
Dependency security	Patched ip-address, sanitize-html, ws, and openai versions are resolved.	Implemented; production npm audit reports zero vulnerabilities
AI policy boundary	Persisted context, channel, and security settings are authoritative; the webview cannot expand them per request. Tool exposure and execution use the same deny-by-default policy.	Implemented and covered by Rust regression tests
Context limits	Requests have message, history, summary, and model-context budgets before a provider call is made.	Implemented; provider billing limits still remain provider-owned
Model discovery	Hugging Face GGUF search and a bounded workspace model-file scan feed the model picker. Only GGUF files are offered to llama.cpp.	Implemented; gated models still require the user's Hugging Face access
llama.cpp	The managed local server can install or use a compatible binary and launch a selected GGUF model. There is no custom Argentum model.	Included; platform-specific binaries and models require explicit setup
Update apply	GitHub release checks are real, but install and rollback actions report unavailable instead of simulating success.	Signed in-place installation is gated on release signing infrastructure

OpenClaw migration behavior

• Argentum scans ~/.openclaw/ after a workspace is selected and from Settings → Migration.
• The backend derives every source and destination path from a fixed item allowlist; the interface
  cannot request arbitrary filesystem copies.
• Existing destination files are retained. An identical re-import succeeds; a different file reports
  a conflict instead of overwriting local data.
• Bundled skills are merged into <workspace>/skills/.
• SOUL.md, MEMORY.md, USER.md, AGENTS.md, and other workspace files are copied under
  <workspace>/imported/openclaw/ for review.
• Database, Telegram credentials, and legacy configuration are copied under
  <workspace>/data/migration/openclaw/ for review. Argentum does not automatically apply them.

Hermes migration remains planned for v0.1.0 and is not offered as a working v0.0.9 action.

Provider Status

Release-candidate providers

ChatGPT/OpenAI, MiniMax, and the managed llama.cpp local-server route are release candidates. They have
working adapters, provider tests, and usage/error surfaces, but the label does not claim that every
credential, quota, model, attachment, or tool combination has been live-tested.

Testing Providers

Anthropic Claude, Google Gemini, OpenRouter, NVIDIA, Groq, external local endpoints, and custom
endpoints remain in testing status.

Supported OS

Platform	Supported release target
Windows	Windows 10/11 x64
Linux	Ubuntu 22.04+ and comparable distributions with WebKitGTK 4.1
macOS	macOS 10.15+ on Intel or Apple Silicon
Android	Android 8.0+ (API 26+)

Hardware Requirements

• Minimum: 64-bit CPU, 4 GB RAM, 500 MB free disk space, and a supported system webview.
• Recommended: 4-core CPU, 8 GB RAM, 1 GB free disk space, and SSD-backed workspace storage.
• Hosted providers require network access and separately configured credentials or account authorization.
• Local GGUF execution depends on model size and available RAM; a discrete GPU is optional.

Planned after v0.0.9

• Translate visible desktop strings through the localization layer.
• Turn Router Agent decisions into real multi-agent session/workspace dispatch with isolation tests.
• Consolidate the remaining knowledge-graph implementations behind GraphBackend.
• Implement and test Hermes migration semantics.
• Decide whether document-generation skills should be distributed with Argentum; upstream catalog
  entries are not represented as bundled capabilities.