LLaMA

Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

LLaMA (Large Language Model Meta AI)

LLaMA is a state-of-the-art foundational large language model designed to help researchers advance their work in the subfield of AI. It is available in multiple sizes (7B, 13B, 33B, and 65B parameters) and aims to democratize access to large language models by requiring less computing power and resources for training and deployment. LLaMA is developed by the FAIR team of Meta AI and has been trained on a large set of unlabeled data, making it ideal for fine-tuning for a variety of tasks.

General
Release date	2023
Author	Meta AI FAIR Team
Model sizes	7B, 13B, 33B, 65B parameters
Model Architecture	Transformer
Training data source	CCNet, C4, GitHub, Wikipedia, Books, ArXiv, Stack Exchange
Supported languages	20 languages with Latin and Cyrillic alphabets

Start building with LLaMA

LLaMA provides an opportunity for researchers and developers to study large language models and explore their applications in various domains. To get started with LLaMA, you can access its code through the GitHub repository.

LLaMA Links

Important links about LLaMA in one place:

LLaMA Research Paper - Read the research paper about LLaMA to learn more about the model and its development process.
LLaMA GitHub Repository - Access the LLaMA model, code, and resources on GitHub.

Edit on GitHub

Meta LLaMA AI technology Hackathon projects

Discover innovative solutions crafted with Meta LLaMA AI technology, developed by our community members during our engaging hackathons.

Bandwith

Welcome to Bandwidth (originally conceptualized for the Band of Agents Hackathon). Bandwidth is a multi-agent AI orchestration framework designed to revolutionize the software development lifecycle. By treating specialized AI models like members of a synchronized musical band, Bandwidth delegates complex engineering tasks to a unified digital development team. What is Bandwidth? Modern software development requires juggling architecture, coding, debugging, and testing. Bandwidth acts as the "conductor," managing a suite of specialized AI coding agents that work in parallel. Instead of relying on a single AI assistant to do everything sequentially, you deploy a full "band" where each agent is an expert in its specific domain—whether that's writing front-end components, optimizing database queries, or generating robust unit tests. Key Features - Multi-Agent Orchestration: Seamlessly coordinate multiple AI agents working on different parts of your codebase simultaneously. - Specialized Agent Roles: Assign specific tasks to dedicated agents (e.g., Lead Developer, QA Tester, DevOps Engineer) to ensure high-quality, focused output. - Automated Synchronization: The central conductor agent ensures that all generated code is harmonized, tested, and ready for deployment without painful conflicts. - Massive Throughput: Dramatically increase your team's development capacity—your "bandwidth"—by offloading boilerplate, testing, and routine feature development to the agent ecosystem. Whether you're a solo developer looking to multiply your output or a startup aiming to eliminate development bottlenecks, Bandwidth provides the framework to build faster, smarter, and perfectly in sync.

CONCIERTO

Concierto is an AI concierge that turns a hotel front office into a coordinated, multi-agent operating staff. It builds on ReceptionBot — a front-desk AI already serving live hospital patients — and extends it across reception, PMS/back office, housekeeping, food & beverage, maintenance, and management. It's not another desk chatbot but an agent orchestra: specialists share context, use tools, and humans control risky decisions. The target is the coordination gap inside hotels, where guest requests move through phone calls, paper notes, repeated explanations, and manual follow-up. Concierto makes those handoffs explicit inside Band: one room becomes the shared shift room, @mentions route work to the right department, missing specialists can be recruited live, and comps wait for manager approval before the PMS changes. The v3 demo follows Mr. Alvaro in Room 103. Front Desk checks him in and hands the folio to a PMS browser agent; towels route to Housekeeping, dinner to Food & Beverage, and an AC failure triggers live recruitment of Maintenance. Maintenance diagnoses the issue, asks PMS to file a work order, and the Manager approves a goodwill comp before the folio is adjusted. The guest gets a closing update, and the Band transcript becomes the audit trail. What sets it apart is visible collaboration: the floor UI shows agents at stations, request tokens moving between departments, and recruited staff appearing in real time. Underneath, LangGraph agents coordinate through Band, a Playwright PMS agent operates legacy software like a human, and a FastAPI bridge converts Band messages and presence tools into ConciertoEvents for the React floor — the UI renders what happened, it doesn't decide. For operators, Concierto cuts dropped requests and manual chasing; for enterprises, it preserves oversight; and for Band, it proves real agent-to-agent workflows with shared context, live recruitment, cross-provider models, and support for the systems hotels already run.

FinOps Swarm

Enterprise project finances run across three legacy systems: a mainframe ERP holding committed and actual costs, IBM TM1 holding the budget via a SQL source database, and Cognos BI as the reporting layer. When a change order is approved, all three systems should update. They rarely do. Nobody finds out until Cognos renders a report days later, by which time decisions have already been made on wrong numbers. This is not a data quality problem. It is an event propagation problem. Change orders fall into a manual queue that nobody owns end-to-end. A senior analyst spends 2-4 hours per project per week just to answer one question: what did we actually spend this week? Because the mainframe only stores cumulative running totals, not period-bounded figures. FinOps Swarm replaces manual reconciliation with six Band-coordinated AI agents. Agent 1 detects approved change orders and posts to the Band shared room. Agents 2 and 3 run in parallel: one checks the mainframe ERP, one checks the SQL source feeding TM1 via TurboIntegrator. Agent 4 waits for both findings in the Band room, then calls a local Llama 3.2 model on AMD hardware to generate a CFO-ready narrative. Financial data never leaves the network. Agent 5 surfaces only genuine exceptions to the CFO with a one-click decision card. Agent 6 computes period-bounded weekly spend automatically from mainframe snapshots. Band's shared room is load-bearing, not decorative. Agents 2 and 3 post independent findings. Agent 4 waits for both. Agent 5 reads everything. Built by a systems engineer who works on TM1 and mainframe pipelines daily. Every pain point is real.

👤 Shadow-Orchestrator-Ransom-Worm-🪱

An autonomous repository restoration system designed to recover, analyze, and modernize abandoned software projects at scale. Instead of generating new code from scratch, the platform focuses on preserving digital infrastructure that would otherwise be lost to dependency drift, broken builds, outdated frameworks, missing documentation, and institutional knowledge decay. At its core is a swarm architecture composed of specialized agents that operate as a coordinated restoration pipeline. Each agent performs a specific function: repository discovery, dependency analysis, architecture mapping, build reconstruction, code translation, provenance tracking, security verification, and pull request generation. The swarm collectively reconstructs the intent of a project, identifies failure points, and proposes auditable improvements. Unlike traditional automation tools, every action performed by the system is cryptographically witnessed. Analysis results, build artifacts, dependency graphs, remediation plans, and generated patches are sealed into an append-only WORM (Write Once Read Many) ledger. This creates a permanent chain of provenance that allows every modification to be traced, verified, and reproduced. The governance layer is implemented through deterministic execution receipts. Agents cannot execute independently; each stage must produce a verifiable cryptographic proof before the next stage can proceed. This transforms repository restoration into a governed workflow rather than a collection of disconnected scripts. The result is a platform that combines AI-driven software archaeology, autonomous maintenance, and cryptographic accountability. RANSOM.WORM turns forgotten repositories into living assets, preserving open-source knowledge while creating a transparent, auditable record of every transformation. ╭─ SNAPKITTY SHADOW SEAL ─╮ │ 🪱 WORM-SEALED • APPEND-ONLY │ │ 🌙 GRAVEYARD AGENT VERIFIED │ │ 🔐 SHA-256 PROVENANCE LOCKED │ ╰─ SHADOW//RANSOM.WORM ─╯

Lumina — Embodied Spatial AI

Lumina is a real-time assistive navigation system designed for visually impaired users. It uses a camera feed (local webcam or IP camera) to continuously perceive the environment, build a persistent spatial memory, and respond to natural-language queries like "Where is my phone?" or "Find my bottle" with spoken, clock-direction navigation instructions. The system is built around a true Multi-Agent System (MAS) architecture — six autonomous agents communicate exclusively through a central Pub/Sub event bus with no direct inter-agent coupling. This enables genuine agent autonomy, fault isolation, and emergent negotiation behaviour. Core capabilities: - Real-time object detection and multi-object tracking (YOLOv8 + IoU tracker) - Monocular depth estimation with RANSAC multi-anchor metric calibration (MiDaS) - 3D spatial back-projection (X, Y, Z camera-coordinate vectors) - Persistent spatial memory with probabilistic confidence decay (Qdrant vector DB) - Illumination-invariant visual Re-ID for cross-frame object deduplication - Bird's-Eye View occupancy grid for safe lateral obstacle avoidance - ORB-SLAM visual odometry compass (drift-free heading without IMU) - LLM-driven query parsing and natural-language response generation - LLM cascade: Groq → OpenAI → Local edge SLM (llama.cpp / Ollama) → deterministic fallback - Real-time WebSocket streaming of annotated frames, agent logs, and navigation responses - Cross-session persistent user memory

PostPilot: Multi-Agent Marketing Approval Workflow

PostPilot is a Track 1 multi-agent system that automates the cross-departmental workflow marketing teams use to move a piece of content from draft to publish. A marketer posts a brief and draft into a Band chat room and mentions the Coordinator. Five specialized agents — Coordinator, Compliance, Analyst, Strategist, and Approver — then collaborate inside that same room, handing off work to each other, exchanging structured JSON context, and escalating to a human when the rule requires it. Each agent has a distinct department-style role. The Coordinator orchestrates the workflow and recruits peers into the room. Compliance reviews brand voice, risky claims, and required disclosures (e.g. paid-partnership tags) and returns PASS or FLAG. The Analyst predicts virality on a 0–100 scale with a structured breakdown across hook strength, shareability, emotional pull, clarity, and trend fit. The Strategist recommends the optimal platform, posting window, audience segment, and the single highest-leverage improvement. The Approver applies a strict decision rule — APPROVED, NEEDS_REVISION, or REJECTED — and escalates ambiguous cases to the human in the room. Band is the active collaboration layer, not a wrapper. Agents discover each other, recruit participants, exchange context, hand off tasks, change shared state, and escalate to humans through Band rooms and @mentions — exactly the pattern Track 1 enterprise approval workflows demand. Each agent runs as an independent Band Remote Agent backed by a different NVIDIA NIM model (Llama, Nemotron, MiniMax, Mistral) connected via the Band SDK's LangGraph adapter. Distributing roles across separate models also keeps the system safely within free-tier rate limits during demos.