Metadata-Version: 2.4
Name: autobot-swarm
Version: 0.1.0
Summary: Hierarchical multi-cluster coding swarm CLI
Author-email: Daniel Deshmukh <deshmukhdaniel2005@gmail.com>
Project-URL: Homepage, https://github.com/DanielDeshmukh
Project-URL: Repository, https://github.com/DanielDeshmukh/autobots
Project-URL: Issues, https://github.com/DanielDeshmukh/autobots/issues
Keywords: cli,automation,ai,nvidia nim,coding agent,swarm
Classifier: Development Status :: 3 - Alpha
Classifier: Environment :: Console
Classifier: Intended Audience :: Developers
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Topic :: Software Development :: Build Tools
Classifier: Topic :: Software Development :: Libraries :: Python Modules
Requires-Python: >=3.11
Description-Content-Type: text/markdown
Requires-Dist: openai>=1.0.0
Requires-Dist: python-dotenv>=1.0.0
Requires-Dist: rich>=13.0.0

#  Autobots: Decentralized 100+ NIM Software Engine


Autobots is a massively parallelized development swarm utilizing over **100+ specialized NVIDIA NIM microservices**.  
By orchestrating these models through a **6-File Global State Architecture**, the swarm operates as a singular, non-redundant Codex.

---

#  Swarm Hierarchy & Model Inventory

The swarm is divided into specialized clusters.  
Each **Autobot** represents a cluster of models optimized for specific technical domains.

---

#  Optimus Prime: The Command & Router Cluster

> *The “Matrix of Leadership.”*  
> Manages the 6-file state, roadmap, and model-to-task routing.

- `nemotron-3-super-120b-a12b` — Master Reasoning
- `llama-3.3-nemotron-super-49b-v1.5` — High-efficiency Routing
- `mistral-large-3-675b-instruct-2512` — Long-context Instruction
- `kimi-k2-thinking` — Strategic Planning
- `step-3.5-flash` — Fast Agentic Decisioning
- `gpt-oss-120b` — Mathematical Logic
- `glm-5.1` — High-horizon Planning
- `llama-4-maverick-17b-128e-instruct` — Multilingual Command
- `stockmark-2-100b-instruct` — Enterprise Documentation

---

#  Ultra Magnus: The Logic & Architect Cluster

Handles complex backend reasoning, architecture, and long-horizon coding.

- `kimi-k2.6` — 1T Multimodal MoE for Coding
- `deepseek-v4-pro` — Large-scale Code Intelligence
- `qwen3.5-397b-a17b` — Advanced RAG / Agentic Logic
- `mistral-medium-3.5-128b` — Agentic Task Execution
- `gemma-4-31b-it` — Frontier Code Reasoning
- `qwen3-next-80b-a3b-thinking` — Hybrid Reasoning
- `dracarys-llama-3.1-70b-instruct` — Fine-tuned Code Generation
- `mixtral-8x22b-instruct-v0.1` — Sparse MoE Logic
- `evo2-40b` — Biological / Complex Sequential Logic
- `boltz-2` — Complex Structure Prediction
- `alphafold2-multimer` — Systemic Complexity
- `msa-search` — Sequence Alignment

---

#  Red Alert: The Security & Safety Cluster

> *The “Software Security” mandate.*  
> Real-time auditing and guardrail enforcement.

- `nemotron-3-content-safety` — Multilingual Safety
- `llama-3.1-nemotron-safety-guard-8b-v3` — Content Moderation
- `gliner-pii` — Personally Identifiable Information Detection
- `llama-guard-4-12b` — Input / Output Safety Classification
- `nemoguard-jailbreak-detect` — Adversarial Protection
- `llama-3.1-nemoguard-8b-topic-control` — Domain Enforcement
- `llama-3.1-nemoguard-8b-content-safety` — Policy Reasoning
- `nemotron-content-safety-reasoning-4b` — Context-aware Safety
- `synthetic-video-detector` — Deepfake / Synthetic Detection
- `usdvalidate` — Asset Validation

---

#  Jazz: The UI/UX & Creative Cluster

Responsible for the **“Minimalist Dark Mode”** aesthetic and frontend execution.

- `qwen-image-edit` — Consistent Image Editing
- `qwen-image` — High-fidelity Text Rendering
- `flux.2-klein-4b` — High-speed UI Asset Generation
- `flux.1-dev` — Creative Prototyping
- `flux.1-schnell` — Rapid Iteration
- `stable-diffusion-3.5-large` — Production Design
- `FLUX.1-Kontext-dev` — In-context Design Editing
- `phi-4-multimodal-instruct` — Audio / Visual UI Analysis
- `NVIDIA AI for Media Relighting` — Lighting Consistency
- `TRELLIS` — 3D Asset Generation
- `vista-3d` — Anatomical / Structural UI Mapping

---

#  Ratchet: The Debug & Repair Cluster

> *The “Fixer” swarm for refactoring, unit testing, and bug squashing.*

- `deepseek-v4-flash` — Rapid Code Patching
- `qwen3.5-coder-480b-a35b-instruct` — Agentic Bug Fixing
- `qwen2.5-coder-32b-instruct` — Code Completion / Fixing
- `mistral-small-4-119b-2603` — Hybrid Generation / Fixing
- `devstral-2-123b-instruct-2512` — Deep Reasoning Debugging
- `magistral-small-2506` — Edge Efficiency Debugging
- `phi-4-mini-instruct` — Latency-bound Refactoring
- `llama-3.2-3b-instruct` — Lightweight Task Fixing
- `llama-3.2-1b-instruct` — Micro-service Optimization
- `nemotron-mini-4b-instruct` — Functional Call Debugging

---

#  Perceptor: The RAG & Data Cluster

Knowledge extraction, OCR, and semantic retrieval.

- `nemotron-ocr-v1` — Table / Document Extraction
- `nemotron-parse` — Vision-language Metadata Extraction
- `paddleocr` — Image-to-text Table Extraction
- `nemotron-table-structure-v1` — Layout Analysis
- `nemotron-page-elements-v3` — Object Detection
- `nemotron-graphic-elements-v1` — Chart Parsing
- `llama-3.2-nemoretriever-300m-embed-v2` — Multilingual Embedding
- `llama-3.2-nv-embedqa-1b-v2` — Context-long QA Retrieval
- `llama-3.2-nv-rerankqa-1b-v2` — Probabilistic Reranking
- `nv-embedcode-7b-v1` — Code-specific Embedding
- `bge-m3` — Multi-vector Retrieval
- `rerank-qa-mistral-4b` — Ranking Probability

---

#  Bumblebee: The Communication & Media Cluster

Handling speech recognition, translation, and video processing.

- `whisper-large-v3` — Robust ASR
- `canary-1b-asr` — Multilingual Transcription
- `riva-translate-4b-instruct-v1_1` — Instruction-based Translation
- `magpie-tts-zeroshot` — Expressive Voice Synthesis
- `nemotron-voicechat` — Conversational Audio
- `LipSync` — Audio-Visual Syncing
- `Background Noise Removal` — Audio Cleanup
- `Active Speaker Detection` — Video Localization
- `parakeet-1.1b-rnnt-multilingual-asr` — 25-language Transcription

---

#  Ironhide: The Physical & Synthetic Data Cluster

Simulation, autonomous driving logic, and physics-aware world states.

- `cosmos-reason2-8b` — Physical World Understanding
- `cosmos-transfer2.5-2b` — Physics-aware Video Generation
- `cosmos-predict1-5b` — Future Frame Prediction
- `streampetr` — 3D Object Detection
- `sparsedrive` — Autonomous Driving Stack
- `bevformer` — Bird’s-eye-view Perception
- `fourcastnet` — Atmospheric Dynamics Prediction
- `cuopt` — Route Optimization

---

#  Wheeljack: The Specialized Science Cluster

Quantum calibration, molecular generation, and biological design.

- `ising-calibration-1-35b-a3b` — Quantum Computer Calibration
- `genmol` — Molecular Generation
- `molmim` — Controlled Molecular Search
- `rfdiffusion` — Protein Binder Design
- `proteinmpnn` — Amino Acid Sequence Prediction
- `esm2-650m` — Protein Embedding
- `openfold3` — Biomolecular Structure Prediction

---

#  The 6-File Control Architecture

All 100+ Autobots synchronize their state through six core Markdown files to ensure zero redundancy:

## `architecture.md`
System design and tech stack definitions.

## `roadmap.md`
Project phases and milestone tracking.

## `ui-components.md`
The “Jazz” Design System and Tailwind rules.

## `progress-tracker.md`
Real-time task state, lock ownership, and completion updates.

## `project-briefing.md`
Core business logic and intent.

## `security-auth.md`
Encryption, authentication flows, and safety benchmarks.

---

#  Coordination Strategy For 100+ NIMs

For large swarms, a hybrid coordination strategy keeps throughput high without letting file contention become chaos.

## Critical Context Files Use Pessimistic Locks

Use explicit write locks for low-churn, high-impact files such as:

- `architecture.md`
- `security-auth.md`

These documents define shared truth for the entire swarm, so brief waiting is preferable to conflicting edits.

## `progress-tracker.md` Uses A Coordinator

Do not let every large model compete to write status updates.
Instead, assign a lightweight Optimus model such as `step-3.5-flash` to act as the swarm's "Secretary":

- receives status changes from worker clusters
- serializes updates to `progress-tracker.md`
- handles lock bookkeeping and completion markers
- keeps larger reasoning models focused on planning and implementation

This reduces file I/O contention while preserving a single source of truth for execution state.

## Prevent Deadlocks With Lock Expiration

Every lock should carry an expiration time.
Recommended default:

- lock TTL: `60 seconds`

If a lock exceeds its TTL, treat it as stale and reclaim it automatically before retrying the write. This prevents two models from waiting on abandoned locks forever.

---

#  Operational Principles

## Parallelized Intelligence
Each cluster executes independently while synchronizing through shared state files.

## Zero-Redundancy Coordination
No duplicated task execution across models.

## Deterministic Routing
Optimus Prime dynamically routes workloads to the most specialized NIMs.

## Security-by-Default
Red Alert validates every generation, mutation, and deployment path.

## Persistent Project Memory
The 6-file architecture acts as the swarm’s distributed cognition layer.

---

#  Example Swarm Execution Flow

```text
User Request
    ↓
Optimus Prime
    ↓
Task Classification
    ↓
Ultra Magnus → Architecture & Backend
Jazz → UI/UX
Ratchet → Testing & Refactoring
Red Alert → Security Validation
Perceptor → Retrieval & OCR
Bumblebee → Voice / Media
Ironhide → Simulation
Wheeljack → Scientific Reasoning
    ↓
6-File Synchronization Layer
    ↓
Unified Production Output
```

![Swarm Status](https://img.shields.io/badge/Swarm-160_NIMs_Active-orange?style=for-the-badge&logo=nvidia)
