Adler AI Workbench Documentation
Adler is a private, low-latency AI developer environment built for rapid coding, reasoning, and automated system engineering with zero retention on model inference.
adler.kitsunetechnologies.org.
1. Quickstart
You can use Adler via three official routes:
- Native Desktop App: Download the Windows executable for tray minimization, background hotkeys, and native notifications.
- Web Workbench: Direct browser access at adler.kitsunetechnologies.org.
- Android Companion: Mobile Compose client with biometric fingerprint/face authentication.
2. Keyboard Shortcuts
| Shortcut | Action | Platform |
|---|---|---|
Ctrl+Shift+Space | Global Show / Hide Summon Adler | Windows Desktop |
Ctrl+K | Focus Thread Search / Filter | Desktop & Web |
Ctrl+Enter or Enter | Send Message / Run Prompt | All Platforms |
Shift+Enter | Insert Newline in Composer | All Platforms |
F5 / Ctrl+R | Hard Reload App Cache | Desktop & Web |
3. Slash Commands
Type any slash command directly in the message composer:
/clear: Instantly starts a fresh conversation thread./export: Triggers direct Markdown (.md) download of the current thread./swift: Switches active engine to Adler Swift ($5/mo tier, sub-300ms speed)./r1: Switches to DeepSeek R1 Chain-of-Thought Reasoner./v3: Switches to DeepSeek V3 Architecture & Coding engine./help: Displays interactive shortcut and command reference.
4. Fulcrum Decision Engine & Adaptive Learning
Adler is paired with Fulcrum, a sub-millisecond deterministic typed decision engine. Fulcrum provides two core capabilities:
A. Dynamic Model Routing (Auto Mode)
When you select ⚡ Auto or send a query without a forced model, Fulcrum classifies the query in <0.1ms:
- Math, proofs, logic, puzzles → Routed to DeepSeek R1 Reasoner with collapsible thought accordions.
- Complex software, refactoring, systems → Routed to DeepSeek V3 Coder with syntax verification.
- Quick questions, definitions, conversational lookups → Routed to Adler Swift (sub-300ms, saves 85% tokens).
B. Personality & Workflow Calibration
Fulcrum learns from your responses and coding patterns. Periodically, Fulcrum poses single-question calibration prompts (e.g. "Do you prefer inline code diffs or complete replacement files?"). Your answers are saved locally in the preference matrix to personalize all future completions without token bloat.
5. Token & Usage Economics
Adler provides real-time token and budget visibility directly in the user panel and system prompt:
| Plan Tier | Price | Daily Quota | Max Context | Economics Focus |
|---|---|---|---|---|
| Community Free | $0/mo | 10 queries/day | 8k tokens | Standard baseline |
| Adler Swift | $5/mo | 100 fast queries/day | 16k tokens | Micro-routed budget ($0.80/M tokens) |
| Adler Pro | $19/mo | Unlimited high-speed | 32k tokens | Dedicated isolated fleet |
| Adler Team | $99/seat/mo | Unlimited + SLA | 64k tokens | Centralized ledger & audit SSO |
6. Token Savings Strategies
Adler assistant actively monitors your token consumption and suggests practical efficiency measures:
- Prune Terminal Dumps: Strip redundant stack traces before pasting. Keep the top 5 stack frames.
- Use Swift for Routine Lookups: Keep heavy reasoning engines for difficult algorithms; run simple syntax and translation queries on Adler Swift.
- Batch Complementary Requests: Instead of 3 back-to-back questions, combine them into one prompt with numbered bullet points.
- Compact Thread History: Use
/clearor start a new thread when switching topics to prevent carrying old context tokens into new queries.
7. Android Mobile Companion
The companion app (Adler-debug.apk) provides a secure mobile portal to your workbench:
- Uses Android BiometricPrompt (fingerprint or face unlock) on device launch.
- Generates non-exportable P-256 keys inside Android's KeyStore hardware module.
- Pairs with your desktop via one-time QR code over Tailscale end-to-end encrypted tunnels.