Octav Launches Comprehensive Web3 API Solution

Octav Labs has released its API as an **Agent Skill**, enabling AI agents to track crypto portfolios, query transactions, and monitor DeFi positions across **65+ blockchains**. **Key features:** - Single command installation: `npx skills add Octav-Labs/octav-api-skill` - Compatible with Claude Code, Codex, ChatGPT, and any agent supporting the open Agent Skills standard - Provides comprehensive DeFi data access through AI agents The skill is available on [GitHub](https://github.com/Octav-Labs/octav-api-skill) and builds on Octav's previous integration with x402scan for DeFi data access.

After repeatedly ignoring user requests, Octav has finally announced they're implementing a highly-requested feature. The announcement came with a humorous tone, acknowledging the platform's initial resistance to user feedback. The team admitted to ignoring multiple requests before ultimately deciding to move forward with the addition. **Key Points:** - Feature was requested multiple times by users - Octav initially dismissed the requests - Platform has now committed to implementation - No specific timeline or feature details provided yet The self-aware announcement suggests Octav is responding to community pressure, though the exact nature of the incoming feature remains undisclosed.

Octav has published seven tutorial videos on their YouTube channel to help new users navigate their portfolio management platform. **What's Covered:** - Adding your first wallet to the platform - Generating complete transaction reports - Essential features for portfolio tracking The tutorials aim to help users master the platform quickly, with step-by-step guidance for common tasks. Octav supports both Ethereum and Solana wallets, with DeFi activity tracking across 25+ chains. [Watch the tutorials](https://www.youtube.com/@octavfi)

Octav fine-tuned their local LLM infrastructure using **Llama Throughput Lab** to maximize performance on Apple M3 Ultra hardware. **Key optimizations:** - Batch and micro-batch sizes - Slot management - Parallelism configuration - Context window efficiency The setup runs **LlamaCPP locally** with GLM-4.7-Flash Q4_K_M (2 instances, 2 parallel slots each) behind an Nginx load balancer. **Why local deployment?** - Removes cloud provider dependency - Enables reliable 24/7 task execution - Provides full control over context and throughput The optimizations achieved maximum speed without crashing Metal, Apple's GPU framework. Tool: [Llama Throughput Lab](https://github.com/alexziskind1/llama-throughput-lab)

**Octav is running LlamaCPP locally on Apple M3 Ultra** with GLM-4.7-Flash Q4_K_M model, using a custom infrastructure setup: - 2 instances with 2 parallel slots each for predictable concurrency - Nginx load balancer distributes requests across instances - Full Metal acceleration utilization keeps latency low **Why go local?** - Removes dependency on cloud providers - Enables reliable 24/7 task execution - Provides complete control over context and throughput This approach prioritizes infrastructure independence and operational control over traditional cloud-based AI deployments.