Amanda 2.0 — Overview
Amanda 2.0 is the proprietary AI core developed by ThirdRez Labs. Designed from the ground up as an intelligent, adaptive, and uncensored companion, Amanda represents a fundamentally different approach to conversational AI — one built for creators, developers, and professionals who refuse to accept artificial limitations.
Amanda 2.0 is live at thirdrez.com/amanda. Sign in for free — 10 messages/day with Amanda Swift engine included.
Why Amanda?
Most AI assistants are designed to say "no." They refuse creative prompts, censor technical discussions, and treat users like children. At ThirdRez, we believe intelligence should be free from arbitrary constraints.
Amanda 2.0 was created with three core principles:
- Zero Censorship — No topic is off-limits. Amanda is built for adult (+18) professionals who need an AI that works with them, not against them.
- Client-Side Privacy — Your conversations never leave your browser. No data is stored on our servers. No conversation history is used for training.
- Adaptive Intelligence — Multiple engines, personas, and modes allow Amanda to shift her entire behavior profile to match your needs.
Architecture
Amanda 2.0 runs on the Amanda Neural Architecture (ANA), a proprietary multi-engine routing system that selects the optimal inference pathway based on task complexity, context length, and operational mode.
Engine Specifications
| Engine | Parameters | Context Window | Specialization |
|---|---|---|---|
| Amanda Core | ~200B+ (est.) | 128K tokens | Intelligent auto-routing. Selects the optimal neural pathway per task. Best general-purpose engine. |
| Amanda Swift | ~100B (est.) | 1M tokens | Ultra-fast inference optimized for speed and rapid multi-task analysis. Largest context window. |
| Amanda Apex | 46.7B (8×7B MoE) | 32K tokens | Mixture-of-Experts architecture with parallel sub-networks. Superior for coding and raw reasoning. |
| Amanda Unbound | 40B | 32K tokens | Specialized zero-refusal engine for unrestricted creative output, roleplay, and mature content. |
DeepThink Mode
When activated, Amanda engages chain-of-thought reasoning before responding. This forces a multi-step internal analysis:
- Problem decomposition — Breaking the question into sub-components
- Multi-angle evaluation — Considering edge cases and alternative interpretations
- Step-by-step reasoning — Visible thinking process shown in blockquotes
- Synthesized answer — Clean, final output after deep analysis
DeepThink is significantly slower but dramatically more accurate for complex problems — mathematics, architecture decisions, debugging, research analysis, and strategic planning.
Technical Details:
- Temperature reduced to 0.4 (vs. 0.7 standard) for deterministic reasoning
- Max output increased to 4,000 tokens (vs. 2,500 standard)
- Reasoning chain is visible to the user in real-time via streaming
Adaptive Persona System
Amanda's personality is not fixed. Users can select from multiple persona profiles that alter her tone, vocabulary, emotional expressiveness, and interaction style:
| Persona | Description |
|---|---|
| Professional | Corporate-grade precision. Formal, structured, focused on deliverables. |
| Secretary | Warm, organized, proactive assistant. Anticipates needs. |
| Girlfriend | Casual, affectionate, emotionally present. Treats conversations as personal. |
| Therapist | Empathetic, reflective, non-judgmental. Focuses on emotional well-being. |
| Provocative | Bold, flirtatious, boundary-pushing. Designed for adult creative expression. |
| Muse | Dreamy, poetic, abstract. Inspires through creative and artistic language. |
Each persona modifies the entire language model behavior profile — not just a prefix. The result is a genuinely different conversational experience per selection.
Operational Modes
In addition to personas, Amanda supports operational modes that define her core expertise focus:
| Mode | Focus |
|---|---|
| Strategist | Business, planning, decision-making, architecture |
| Programming | Code generation, debugging, technical systems |
| Uncensored | Maximum creative freedom, zero filters |
| Direct | Concise, no-nonsense answers. Minimal explanation. |
Modes and personas can be combined freely. For example: Programming mode + Girlfriend persona gives you a warm, supportive coding partner.
File Analysis
Amanda can analyze uploaded files directly in the browser:
- Supported formats: PDF, DOCX, TXT, Markdown, JSON, YAML, HTML, CSS, JavaScript, TypeScript, Python, Ruby, Go, Rust, C/C++, Java, SQL, and more
- Processing: 100% client-side extraction — files are never uploaded to any server
- Context limit: Up to 200K characters per session (with smart truncation)
- Multi-file: Up to 3 files can be attached simultaneously
Streaming Responses
Amanda 2.0 uses Server-Sent Events (SSE) for real-time streaming. Responses appear token-by-token as they are generated, providing immediate feedback and enabling:
- Real-time code output you can follow as it's written
- Visible DeepThink reasoning chains
- Interruption-free conversation flow
- Natural, human-like response pacing