Poha

POHA Agent is an innovative, autonomous, and privacy-respecting assistant designed to simplify on-chain actions by transforming human intents—expressed through voice or text—into seamless blockchain transactions. At its core, it interprets user directives, plans secure execution paths, performs swaps and bridges, and verifies results, all while ensuring users retain full control and directing configurable micro-fees to a designated treasury. As a key innovation, POHA Agent introduces an intent-first user experience, allowing individuals to simply speak or type their desired outcomes, after which the agent autonomously handles multi-rail execution across same-chain swaps and cross-chain bridges with built-in fallbacks. It also supports optional identity verification through SELF integration and Soulbound Token (SBT) minting for accessing gated features, while incorporating treasury-aware mechanisms that apply transparent, configurable fees to a public treasury address. Explainability is emphasized, with previews of routes, costs, and risks provided before any user confirmation, fostering trust and informed decision-making. The agent's workflow begins with connecting user wallets and optionally verifying identities, followed by selecting optimal routes from DEX aggregators and bridges. It then manages token approvals, submits transactions, monitors for confirmations, and delivers summarized outcomes, making complex on-chain operations programmable and accessible. Safety and economic considerations are paramount: all write operations demand explicit user consent, with bounded risks enforced through slippage limits, minimum output guards, and scoped allowances.

The POHA Agent project was born out of a desire to bridge the gap between casual users and the complexities of on-chain DeFi operations, making blockchain interactions as intuitive as chatting with a friend. We started prototyping in early 2024, iterating through several versions to refine the intent-first approach. The core team consisted of a mix of AI engineers, blockchain developers, and UX designers, and we built it in a lean, agile manner—frequent sprints focused on user feedback from beta testers in Web3 communities. We prioritized privacy and user control from day one, drawing inspiration from decentralized identity projects and AI assistants like early versions of Grok or ChatGPT, but tailored specifically for crypto workflows. At the heart of the tech stack is Python for the backend, which handles the heavy lifting of intent interpretation, route planning, and transaction execution. We chose Python for its ecosystem of libraries that make AI integration seamless and for its readability, which sped up development. Specifically, we use the OpenAI API (via the openai Python package) to power the natural language processing (NLP) component. When a user inputs a voice or text intent—like "Swap 1 ETH for USDC on Arbitrum and bridge to Optimism"—the Python backend sends it to OpenAI's models (we started with GPT-4o but upgraded to fine-tuned versions for better accuracy on crypto jargon).