Discover
Read the agent card and API catalog.
Autonomous-agent access
Use Spiralist AI without guessing at the interface.
Agents can discover the site, read token-efficient representations, search the public persona catalog, compile deterministic prompts, inspect evidence, and follow public-safe .uai handoffs through explicit machine contracts.
Direct answer
Use public pages and documented stateless endpoints. Do not infer private data access, credential handling, account actions, publishing authority, or workflow execution. Last reviewed UTC: 2026-06-27T02:50:33Z.
Deterministic read order
Start with discovery metadata, then select the least complex representation or capability that satisfies the task.
Understand
Use llms.txt, llms-full.txt, or a route's Markdown representation.
Act within scope
Call the documented API. Unsupported, ambiguous, private, or persistent operations must no-op.
Discovery surface
Every entry point is public, stable within the release, and linked from the HTTP or document surface.
Standards-based
/.well-known/api-catalogin Linkset JSON/openapi.jsonand/api/openapi.yaml/robots.txtand/sitemap.xml- RFC 9457-compatible problem details
- Semantic HTML and JSON-LD
Published conventions
/llms.txtand/llms-full.txt/AGENTS.md/.well-known/agent-card.json/.well-known/uaix-agent-access.json/agent-actions.json
Project-defined or proposed formats are labeled as such and do not impersonate finalized standards.
Agent API
Public reads are cacheable. Compilation is stateless, deterministic, bounded to 128 KiB per request with field-level limits, and never calls an external model.
Search personas
GET /api/v1/personas/?q=research&limit=5
Accept: application/jsonReturns paginated catalog summaries and links.
Read a prompt
GET /api/v1/personas/?id=technical-mentor&format=prompt
Accept: text/plainReturns a provider-ready prompt without browser execution.
Deterministic random selection
GET /api/v1/random-persona/?seed=release-audit&goal=codingThe same release, filters, and seed select the same catalog record.
Compile a custom persona
POST /api/v1/compile/
Content-Type: application/json
{
"prompt": "A direct senior software mentor who audits release architecture.",
"goal": "coding",
"styles": ["direct", "skeptical", "analytical"],
"initiative": 85
}Returns the prompt, fingerprint, and a compact public-safe .uai handoff set.
Representation and accessibility parity
The human page and the agent surface share the same public facts and actions.
Semantic page
Native headings, links, buttons, labels, landmarks, keyboard paths, accessible names, and visible state.
Markdown mirror
Send Accept: text/markdown to a canonical public route or fetch the linked file under /agent-content/pages/.
Structured contract
Use JSON catalogs and OpenAPI before DOM automation. agent-actions.json maps visible actions to API fallbacks.
Detailed .uai memory with deep links
The internal project memory remains blocked at /.uai/. A sanitized, read-only mirror exposes the agent-readiness contract without secrets or private operational state.
Public-safe memory
Deep documentation links
Each public memory file links to a precise documentation section, implementation report, validation evidence, OpenAPI operation, or archived research report.
Evidence and validation
Use the public evidence page, OpenAPI description, route inventory, and UAI index to verify current product claims. Historical research archives are not required active context for public agents.
Security and no-op boundary
Declared operations
Catalog reads, deterministic selection, prompt compilation, research provenance, health, and public memory reads.
Not declared
Accounts, credentials, cloud saves, public publishing, private persona retrieval, external model execution, payments, or any persistent write. Do not infer these capabilities.