SuperNova Desktop AI Workbench

A Windows-first desktop AI workbench that turns chat into traceable tasks, backed by runtime receipts and replayable evidence for every local execution.

Stars
★ 188
Last updated
4mo ago
License
Apache-2.0
Primary language
Rust

At a glance

Works with
Platform-specific
You'll need
Rust/CargoNode.js and npmTauriDeepSeek provider configuration for live model callsShell / CLINetwork accessLocal filesystem
Typical use
A developer facing an unfamiliar codebase uses Chat to read source, explain module relationships, and inspect diffs, then uses TASK to apply workspace-scoped code changes and record what actually changed
Main limitation
Windows-first delivery: the documented installer is a Windows NSIS .exe, with no documented macOS/Linux path
Maintenance
Commits in the last 6 monthslast commit 2026-07-02

What does this agent do, and when should you use it?

SuperNova (repository Ray-Code-Svg/SuperNovaAgent) is a Windows-first desktop AI Agent workbench that pairs a Tauri + React Workbench v2 front end with a Rust Product Runtime and Process Kernel. It brings project files, Chat, TASK execution, context, and output destinations into one surface, where a chat answer is just text while a TASK is tracked local work with runtime state, artifacts, and completion evidence. A typed Local Runtime Protocol carries DTOs between the UI, the product runtime, and the kernel, while the Process Kernel maintains ChatTruth and ProcessTruth stores of receipts and replayable execution events, projected into the Product DB and streamed back to the UI over SSE. The project is an RC0 stabilization candidate with a committed Windows NSIS installer and a documented build-from-source path using Cargo and npm; live model calls rely on a locally configured DeepSeek provider path. It suits users who want auditable, verifiable agent execution inside a local workspace rather than a hosted-only service.

SuperNova exposes two user-visible modes in Workbench v2. Chat reads and explains workspace content, including files, directory trees, hashes, diffs, datasets, Office text, PDF text, and sanitized local environment facts. TASK performs controlled local execution: the command composer selects mode, sources, and artifact target; a generated protocol client sends typed requests over HTTP routes and SSE; product services start or resume runs through the Run manager and KernelApi into TaskRuntime (RootProcess + TaskAgent). Provider-native tool calls are treated as intent until registered capabilities (Office/OS/terminal/artifact) execute and record receipts. Concrete operations include workspace-scoped file create/edit/copy/move/rename/delete/unzip, bounded foreground commands and start/stop/check of long-running services, DOCX creation and rewrite plus validation via the Office Worker, export of CSV/Markdown/JSON/TXT artifacts with row counts, schema, and derivation facts in receipts, zip packaging with manifest/checksum files, and display of run state, artifact cards, completion statements, and evidence on the task timeline.

  1. A developer facing an unfamiliar codebase uses Chat to read source, explain module relationships, and inspect diffs, then uses TASK to apply workspace-scoped code changes and record what actually changed
  2. An office worker processes documents: Chat reads DOCX/PDF/workbook content, TASK creates or rewrites DOCX files and validates the produced document
  3. A data analyst reads CSV datasets and exports CSV or Markdown results, with receipts preserving row counts, schema, and derivation facts
  4. Running dev services: TASK starts, stops, and checks long-running services and attaches terminal results to the task timeline
  5. A user producing deliverables builds zip packages with supporting manifest/checksum files and verifies the package artifact
  6. A developer extending the platform adds TASK flows, registers new capabilities, or builds new Workbench views over the Product Runtime streams

How do you install or deploy this agent?

Option 1: install the committed Windows package. The navigation entrypoint is the releases/windows/ directory; download the latest committed NSIS .exe installer and run it.

Option 2: build from source (requires Rust/Cargo and Node.js/npm):

cargo check --workspace
npm.cmd --prefix desktop_shell/ui run typecheck
npm.cmd --prefix desktop_shell/ui run build
npm.cmd --prefix desktop_shell/ui run tauri:build

The local Tauri build writes the Windows NSIS installer under desktop_shell/src-tauri/target/release/bundle/nsis.

Note: live provider validation requires local provider configuration; do not commit API keys, private screenshots, or implementation-level security details.

How do you use this agent?

After installing and launching the SuperNova desktop app, pick a local workspace or container in Workbench v2 and use the slash-command composer to switch between Chat and TASK modes, select sources (workspace files, folders, Chat and TASK history), and set an artifact target. The model flyout configures provider, model, thinking, reasoning effort, and output budget; the context flyout includes recent Chat turns, TASK runs, summaries, and explicit context items. Use Chat for questions; switch to TASK when you need actual changes, commands, or artifacts, and watch the active run, task stream, status, artifact cards, and completion statement in the Workbench. Configure the DeepSeek provider path locally before making live model calls.

What are this agent's strengths and limitations?

Pros
  • Truth separated from intent: model tool calls are only intent until registered capabilities record receipts, avoiding the unverifiable output typical of chat-only agents
  • Typed protocol and layered architecture: Local Runtime Protocol DTOs, Product Runtime projections, and Process Kernel boundaries let extensions pick the shallowest layer while keeping receipt-backed evidence
  • Local-first desktop execution: workspace, run state, artifacts, and evidence live inside the local Workbench, suited to auditable, replayable long-running tasks
Limitations
  • Windows-first delivery: the documented installer is a Windows NSIS .exe, with no documented macOS/Linux path
  • Pre-release status: RC0 candidate, not a final release; the repo requires fresh verification for current-state claims
  • Narrow provider path: the documented model runtime path is DeepSeek; other model providers are not documented
  • Building from source requires a full Rust + Node.js/Tauri toolchain, a real hurdle for non-developers

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
SuperNova Desktop AI Workbench This agent 59 · Major gaps — ★ 188 4mo ago Rust —
Claudette Workspace Orchestrator 70 · Some gaps Desktop app ★ 88 13d ago TypeScript Claude Code
Hermes Agent CN Desktop 55 · Major gaps Desktop appFree + model costs ★ 1.9k 1d ago TypeScript —
OpenAgentd 53 · Major gaps Desktop appFree + model costs ★ 217 2d ago TypeScript ChatGPT · OpenAI API · Claude API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
59/ 100 5-point scale 3.0 / 5
Trust 16/29
Reliability 8/14
Adaptability 10/18
Convention 10/18
Effectiveness 9/13
Verifiability 6/8
Why each dimension lost points
Trust16 / 29 · 2.8/5

Positives: default bind 127.0.0.1; ApprovalRuntime, CapabilityApprovalGuard, closure_gate, redaction-report modules exported; docs require tool execution to be tied to receipts. Deductions: main.rs sets runtime_token None and empty allowed_origins, so loopback auth is weak in evidence; approval/closure logic appears only as exported symbols with no visible implementation, so 2; rollback shows only replayable events with no user-facing rollback mechanism (1); dependency_security lacks lockfile/audit evidence (1); source_attribution has no author/maintainer info (1).

Reliability8 / 14 · 2.9/5

Positives: error envelope includes code/status/retryable; kernel schema version constants are consistent and naming is stable. Deductions: dependency availability is only a prerequisites list without pinned versions (1); key execution paths (TaskAgent, approval enforcement) are not in the packet, so full marks for self-consistency are not earned.

Adaptability10 / 18 · 2.8/5

Positives: README separates everyday users and developers with scenario tables; environment fit is explicit (Windows-first, PowerShell, Tauri). Deductions: trigger precision is only supported by slash-command screenshots with no textual trigger rules (1); capability boundaries are described rather than enforced in visible code, capped at 2.

Convention10 / 18 · 2.8/5

Positives: bilingual docs, mermaid architecture, layered extension tables; install notes cover installer entry and source build; README and quickstart agree; known limitations state RC0/non-final; complete Apache-2.0 LICENSE consistent with Cargo.toml. Deductions: examples/FAQ are screenshots only with no FAQ or reproducible examples (1); no CHANGELOG/version history (1); no CONTRIBUTING or maintainer commitment (1).

Effectiveness9 / 13 · 3.5/5

Positives: outputs include zips with manifest/checksums, CSV/Markdown/DOCX artifacts, receipts and artifact cards; truth-backed local execution gives clear marginal value over plain chat; local runtime plus codegen client aids cost control. Deductions: all benefit claims are self-described without quantitative evidence, so not full marks.

Verifiability6 / 8 · 3.8/5

Positives: README explicitly says screenshots are not validation evidence and release claims require fresh verification — excellent fact/inference separation (3); architecture diagram, schema codegen, and docs corroborate each other adequately (2). Deductions: claim traceability rests on the receipts design with no visible evidence chain in implementation (2).

Risks and how to mitigate them
  • Static review only: critical security paths (approval/closure gate, terminal execution) appear as exported modules without visible implementation; audit before use.
  • product_runtime defaults to runtime_token None and empty allowed_origins; the auth strength of the local HTTP/SSE interface is unverified.
  • No changelog, maintainer information, or dependency-lock evidence in the packet; the distribution is a NSIS .exe committed in-repo — verify provenance and checksums before installing.
  • The product self-identifies as RC0, not a final release; no capability claims were verified by this review.
Evidence confidence: Low Reviewed Nov 22, 2026 Reviewed revision 13b334c72364
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/supernovaagent.svg)](https://followagents.com/en/agents/supernovaagent)
<a href="https://followagents.com/en/agents/supernovaagent"><img src="https://followagents.com/badge/supernovaagent.svg" alt="FARS review"></a>

FAQ

Is SuperNova Desktop AI Workbench open source?
Yes. The source code is on GitHub (Ray-Code-Svg/SuperNovaAgent) under the Apache-2.0 licence.
Is SuperNova Desktop AI Workbench safe to use?
FollowAgents' FARS review (Nov 22, 2026) scored SuperNova Desktop AI Workbench 16/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is SuperNova Desktop AI Workbench still maintained?
Development has slowed: the latest commit on GitHub is from 2026-07-02, within the last six months but not the last 30 days.
Who makes SuperNova Desktop AI Workbench?
SuperNova Desktop AI Workbench is published on GitHub by the account Ray-Code-Svg. FollowAgents has not verified which person or company is behind that account.
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts