Access External LLM and Image Capabilities

BlockRun is a skill for Claude Code and Google Antigravity that gives an agent capabilities it otherwise lacks - image generation and real-time X/Twitter access - by paying for them autonomously through a wallet. Instead of API keys, BlockRun uses x402 micropayments to route requests to OpenAI, xAI, Google, and other providers, with the wallet paying per token or per call: image generation via DALL-E ($0.04/image), real-time X data via Grok with Live Search ($0.025/source), a second opinion from GPT-5.2 ($1.75/M input, $14/M output tokens), or cheaper bulk processing via DeepSeek ($0.14/M input, $0.28/M output).

Use it when the user explicitly asks for it ("blockrun second opinion with GPT", "use grok to check...", "generate image with dall-e"), or when the agent needs something it genuinely can't do itself, like generating an image or pulling live X/Twitter data - in that case, suggest BlockRun and wait for confirmation rather than invoking it silently. If the agent can already handle the task itself, it should just do it and not mention BlockRun. An optional budget control tracks cumulative spend via get_spending() and stops making calls once a user-specified budget (e.g. "use at most $1") is reached.

Every session starts with setup_agent_wallet(), which auto-creates a wallet (showing a QR code on first use) and returns a client:

from blockrun_llm import setup_agent_wallet

client = setup_agent_wallet()  # Auto-creates wallet if needed
response = client.chat("openai/gpt-5.2", "What is 2+2?")
print(response)

### Check spending
spending = client.get_spending()
print(f"Spent ${spending['total_usd']:.4f}")

Real-time X/Twitter data requires explicitly enabling Live Search with search=True, or search_parameters for advanced filtering (mode, sources of type x/web/news with handle/domain allow-lists, max_search_results to control cost, return_citations, date ranges). Image generation uses a separate ImageClient().generate(prompt), returning a result with a .url. Nine models are available spanning reasoning (openai/o3, openai/o4-mini), cost-optimized chat (gpt-5-mini, deepseek/deepseek-chat), long-document handling (google/gemini-2.5-flash), real-time search (xai/grok-3), and image generation (openai/dall-e-3, google/nano-banana), each with its own per-token or per-image pricing.

Installed via pip install blockrun-llm, with the wallet session stored at $HOME/.blockrun/.session and funded with $1-5 USDC on the Base network (a QR code can be generated in-terminal via generate_wallet_qr_ascii). Wallet balance (get_balance(), on-chain USDC) and address (get_wallet_address()) are checkable at any time. A common troubleshooting case is Grok reporting no real-time access - the fix is adding search=True to the call - and a "module not found" error resolves with the same pip install blockrun-llm command, upgraded via pip install --upgrade blockrun-llm.

Agents and their developers working in Claude Code or Google Antigravity who need occasional access to capabilities outside the base model - image generation, live social-media data, a second-opinion model, or cheaper bulk processing - paid for autonomously per use rather than through a pre-provisioned API key.