AI coding agents like Claude Code, Codex, OpenCode, Cursor, and so on now write code, provision infrastructure, swap out RPC providers, and ship to prod without a human chiseling each line out of a text editor anymore. But most RPC providers are still requiring humans to click through dashboards to use their services. The result: agents waste cycles scraping docs, guessing at API shapes, and re-discovering information that should be one command away.
We are closing that gap at Dwellir for all of you that like to clanker around in your projects. We call it Agent Experience (AX) - the practice of designing infrastructure, documentation, and tooling so that AI agents can consume them as efficiently as a senior engineer reads a README. AX is to agents what DX is to developers: the difference between "it technically works" and "Claude Code just one-shotted this".
Today, the Dwellir team is shipping four tools that put AX into practice. You can explore all of them on dwellir.com/agents.
Dwellir CLI
The Dwellir CLI gives agents (and developers) full access to the Dwellir platform from the terminal. Authenticate, manage API keys, discover 140+ blockchain endpoints, monitor usage, and inspect error logs without touching a browser.
Every command returns structured output wrapped in a standard {ok, data, meta} envelope. The CLI detects agent environments automatically by checking for markers like CLAUDECODE, CODEX_THREAD_ID, CURSOR_AGENT, and others. When running inside an agent, the CLI defaults to TOON output, a compact structured format that agents parse with ~24% fewer tool calls and ~37% fewer output tokens compared to JSON, based on our internal benchmarks across Claude and Codex runs. You can still use --json or --human to override.
Our local Codex/Claude benchmark run showed improved agent ergonomics with TOON (equal correctness, fewer tool calls, and lower output-token footprint): Agent output benchmark (2026-03-03).
For TOON trade-offs and broader benchmarks, see:
Per-project authentication via CLI Tokens is also supported. If you are working on both personal and work projects and use Dwellir for both, you can scope credentials to each project instead of sharing a single global session.
curl -fsSL https://raw.githubusercontent.com/dwellir-public/cli/main/scripts/install.sh | sh
# Authenticate
dwellir auth login
# Discover Ethereum endpoints (TOON output in agent environments, JSON with --json)
dwellir endpoints search ethereum
# Create an API key
dwellir keys create --name my-agent-key
# Check usage and error logs
dwellir usage summary
dwellir logs errors --status-code 429
The CLI is also available via Homebrew, AUR, go install, and from source.
Agent-Optimized Documentation
Dwellir's documentation is structured so agents can discover and consume content without parsing HTML.
llms.txt - The site follows the llmstxt.org specification. Agents fetch a single file and get a structured index of every available page:
| Endpoint | What it returns |
|---|---|
/llms.txt | Site-wide index covering all networks, blog posts, and docs |
/docs/llms.txt | Curated documentation index with network guides |
/docs/llms-full.txt | Every doc page concatenated into one file for large-context agents |
Markdown endpoints - Use the dedicated docs API endpoint or the existing .md shortcuts on networks and blog pages to get clean markdown with Content-Type: text/markdown headers. All JSX/MDX components are stripped, internal links are rewritten to absolute URLs, and YAML frontmatter is preserved:
https://www.dwellir.com/api/docs/ethereum/markdown
https://www.dwellir.com/networks/ethereum.md
https://www.dwellir.com/blog/hyperliquid-rpc-providers-2025.md
Smart 404s - When a page is not found, the API returns fuzzy-matched suggestions ranked by confidence instead of a generic error. If there is exactly one high-confidence match, the server redirects automatically. Minor typos and case mismatches resolve without the agent needing to retry.
Content negotiation - All markdown endpoints set Link headers with rel="canonical" and rel="alternate", giving agents a machine-readable way to navigate between HTML and markdown versions of the same page.
For the full list of agent-facing endpoints, see the Agent Tooling docs page.
RPC Migration Prompt
Switching from Alchemy, Infura, QuickNode, or another provider to Dwellir? The migration prompt automates the process. Copy it into your AI agent's context and point the agent at your codebase. It walks through five phases: environment discovery, codebase scanning, compatibility matching, migration, and summary reporting.
The prompt handles common edge cases that manual migrations miss. It resolves chain ID ambiguities across providers, detects archive node requirements by scanning for historical state calls, checks for EVM client-specific response shapes (Geth vs. Reth trace formats), and flags unsupported chains rather than silently breaking them.
You can find the full prompt on the Getting Started page, the Agent Tooling docs, or the Agents hub on the Dashboard. All three include a copy button for easy pasting.
Hyperliquid Agent Skill
The Hyperliquid agent skill gives AI coding agents procedural knowledge for building on Hyperliquid through Dwellir's infrastructure. It follows the open Agent Skills standard, making it portable across 40+ AI coding agents including Claude Code, Cursor, and Windsurf.
The skill is listed on skills.sh and you can easily install it with:
npx skills add dwellir-public/hyperliquid-skills
Once installed, the skill activates automatically when an agent encounters Hyperliquid-related tasks. It covers:
- HyperEVM JSON-RPC - Query EVM state and deploy Solidity contracts (Chain ID: 999)
- Info API - Market data, prices, order books, candles, funding rates, user positions
- gRPC streaming - Real-time L1 block data, fill executions, order book snapshots
- Order book WebSocket - L2 and L4 depth data from edge servers in Singapore and Tokyo
- Read/write routing - The skill teaches agents to route read operations through Dwellir endpoints and write operations (orders, transfers) through Hyperliquid's native API
And more as we build out our Hyperliquid data stack.
We benchmarked Claude Opus 4.6 and OpenAI Codex across four tooling configurations building Hyperliquid trading dashboards. Agents with the skill shipped 70% faster at 57% lower cost, with compounding gains when skill and CLI knowledge were combined. Read the full results in AI Coding Agents Build Better Apps with Infrastructure Skills.
Get Started
All four tools are live today:
- CLI: Install and authenticate in under a minute
- Docs for agents: Fetch
/llms.txtor append.mdto any doc URL - Migration prompt: Copy from Agent Tooling and point your agent at your codebase
- Hyperliquid skill: Run
npx skills add dwellir-public/hyperliquid-skillsin your project
The Agent Toolkit page is the hub for everything above, with the technical documentation covering every endpoint and flag in detail. For questions or feature requests, reach out at support@dwellir.com or join the conversation on Telegram.
