Agent Quickstart
Use this page when an AI agent needs to understand Elfa quickly, choose the right surface, and avoid scraping docs page by page.
Load Context First
Agents should load the high-signal sources before walking individual docs pages:
- Read
llms.txtfor the docs map and canonical machine-readable links. A plainGET https://docs.elfa.ai/returns that same index astext/plainto any client that does not ask for HTML, so an agent holding only the bare domain does not need to know this path in advance. - Read the raw Elfa Agent Skill directly when the environment cannot install skills.
- When skills are available, install or refresh with
npx skills add elfa-ai/skills. - Use the Endpoint manifest or OpenAPI for exact request/response metadata.
- Use
llms-full.txtonly when deeper page context is needed.
Refresh the Elfa skill weekly so agents pick up the latest Auto capabilities and integration guidance as they ship.
Pick an Integration Surface
Context loading and reaching the API are separate choices. Pick how the agent calls Elfa:
| Surface | Use when |
|---|---|
| Agent Skill | The agent has a shell. Costs no context until it is used. |
| MCP Server | The client cannot run commands. Sixteen tools and no integration code. |
| SDKs | You are writing TypeScript or Python. Typed responses, retries, and SSE are handled. |
| REST | Anything else, or a language without an SDK. |
Choose the Right Surface
| If the agent needs to... | Start with | Why |
|---|---|---|
| Know what happened and how a story developed | API v3 Events | Events with their cited posts and measured price reactions, threaded into storylines; searchable by topic. |
| See who is long or short, and whose calls to trust | API v3 Calls | Open positions on an asset, each position from open to close, and the author's track record. |
| Ask a market question or summarize current context | Market Intelligence Chat | Fastest path to written interpretation without building a retrieval pipeline. |
| Discover attention, accounts, tokens, news, or narratives | Market Intelligence | On-demand structured signals and source-backed context. |
| Chain market data calls into a workflow | Market Intelligence Agent Playbooks | Endpoint chains for lead discovery, influence vetting, cause analysis, and evidence gathering. |
| Watch continuously and fire when conditions resolve true | Auto | Managed condition engine, trigger pipeline, delivery actions, and optional LLM analysis. |
| Build an Auto plan end-to-end | Auto Agent Quickstart | Deterministic validate -> create -> stream/poll flow for agents. |
| Continue work after an Auto trigger | Auto Agent Runner | Worker pattern for policy, queueing, replay, and follow-up decisions. |
| Build a custom UI on Elfa's runtime | Widget Headless Quickstart | Server-to-server widget surface with end-user identity, sessions, quotas, and personalization. |
| Connect a client that cannot run commands | MCP Server | Tools for Claude, Cursor, VS Code and other MCP clients, with no integration code. |
| Pay per request without provisioning an API key | x402 Payments | Wallet-native access for autonomous agents. |
Agent Boot Sequence
- Load the skill or raw skill file.
- Read
llms.txt. - Pick the product surface from the table above.
- Use the endpoint manifest or REST reference for exact method, path, auth, cost, and examples.
- Validate requests before mutation where the product supports validation.
- Prefer product-specific quickstarts for execution order and error handling.
Access Model Routing
| Model | Best when | Docs |
|---|---|---|
| API key | You want account-linked usage, plan limits, and monthly or PAYG credits. | Authentication |
| Auto API key | You create, poll, and manage Auto plans with account-linked credits. | Auto API Key Auth |
| x402 | The agent has a wallet and should pay per request without an API key. | x402 Payments |
| x402 Auto | The agent should create, poll, and manage Auto plans with wallet-based payment. | Auto x402 Instructions |
| Widget key | A partner backend calls Elfa for many end-users under its own UI. | Widget Headless Quickstart |
Runtime Data Rules
For Auto runtime data, distinguish:
latestEvaluation.conditionStates[]: latest current-state values.- Auto Trigger Context on executions (
triggerTime,conditionsMet,trigger): what fired. trigger.matchedConditions[].match.observedValue: value observed at trigger time.
For Market Intelligence data, remember:
- Measurement endpoints return counts, links, and account/token metadata.
- Interpretation endpoints return written analysis or summaries.
- Measurement endpoints do not return raw tweet text or sentiment fields.