v0.1.0 DESKTOP AGENT WORKSTATION

Run AI agents with
verifiable control.

Bring your own API keys across 13+ built-in adapters and custom local gateways with automatic failover, diff-first file edits, and OS keychain security.

OS Keychain Storage Diff-First File Edits Approved Tool Execution
VersaRail AI
Harden atomic writer Ready
You 10:41 PM
Harden atomic-writer.ts and verify SHA-256 baseline before commit.
AI Agent 10:41 PM gemini-2.5-flash
src/main/changes/atomic-writer.ts +4 -1
@@ -42,4 +42,7 @@ commitFile(req)
-  await fs.writeFile(path, next)
+  const base = await sha256(path)
+  if (base !== hash) throw Conflict()
+  await writeTempAndFsync(tmp, next)
+  await fs.rename(tmp, path)
Baseline verified
Ask AI Agent to modify code, run tests, or @ mention files...
AI PROVIDERS / DYNAMIC FALLBACK

Connected Endpoints

Pre-First-Byte Failover: Routing primary inference through Gemini 2.5 Flash (184ms). Backup lease: Groq.
Gemini
Google Gemini gemini-2.5-flash · ••••x924
Primary
Groq
GroqCloud llama-3.3-70b · ••••k418
Standby
Cerebras
Cerebras qwen-3-32b · ••••c109
Healthy
Ollama
Ollama Local 127.0.0.1:11434 · offline
Local
SETTINGS / PREFERENCES & PRIVACY

Workspace Configuration

Appearance Theme synced
Privacy & Context Management Auto-Summarize: ON
RETENTION30 DaysLocal SQLite purge
CONTEXT BUDGET32,000Tokens per turn
ALLOWED ADAPTERS13 / 13Scoped per run
USAGE & QUOTAS / TOKEN BUDGET

Rate Limits & Usage

AUTO-SUMMARIZE: ON
Context Window Budget14.2k / 128k
Gemini Daily Free Tier62% remaining
FAILOVERS SAVED02Pre-first-byte
STORAGE ENGINELocalSQLite WAL
24H TOKENS48.6k3 active endpoints
01MULTI-PROVIDER ENGINE

Bring every model. Fail over cleanly.

Connect 13 built-in adapters—plus Ollama or any OpenAI-compatible custom endpoint—straight from your machine. Pre-first-byte rate limits or server errors switch backups automatically; mid-stream errors pause cleanly instead of mixing responses.

Google Gemini
GroqCloud
Cerebras
OpenRouter
DeepSeek
Mistral AI
Cloudflare AI
NVIDIA NIM
Cohere
Hugging Face
Kilo Gateway
Pollinations
Ollama (Custom)
OpenAI-Compatible
Custom Gateway
02CAPABILITIES & SECURITY

Built for verifiable local control.

Every model request, file edit, and command stays transparent and auditable on your machine.

Multi-Provider Failover

Bring 13 built-in adapters plus custom local Ollama and OpenAI-compatible gateways. Rate limits or server errors switch backups before the first token renders.

Diff-First Atomic Edits

Review every proposed change as a unified diff before anything hits disk. File hashes detect external modifications so writes apply cleanly and atomically.

Verified Tool Approvals

File mutations and sandboxed commands—with elevated actions like git push flagged explicitly—are locked to your approved arguments with timeouts, output caps, and secret redaction.

Usage & Quota Tracking

Track 24-hour and monthly token usage, free-tier budgets, and live provider cooldown timers from one workspace view.

OS Keychain Encryption

API keys stay encrypted inside Windows DPAPI, macOS Keychain, or Linux Secret Service—never stored in unencrypted plaintext files.

Local-First & Zero Telemetry

Multi-thread chats, @ file mentions, context budgets, and configurable history purges live in local SQLite inside an offline desktop UI.

Want implementation internals? Architecture Docs ↗ Security Threat Model ↗
03DESKTOP INSTALLERS · GITHUB RELEASES

Install VersaRail AI.

Download desktop builds from GitHub Releases and verify checksums locally.

Verify checksum (PowerShell): Get-FileHash .\VersaRail-AI-Setup-win-x64.exe -Algorithm SHA256

Unsigned early build: verify SHA-256 first, then click More info → Run anyway if SmartScreen prompts.

Verify checksum (Terminal): shasum -a 256 ~/Downloads/VersaRail-AI-mac-arm64.dmg

Unsigned early build: verify SHA-256, then use Right-click → Open on first launch.

Verify checksum (Shell): sha256sum VersaRail-AI-linux-x86_64.AppImage

Requires an active desktop keyring (GNOME Keyring or KWallet); plaintext fallback is refused.