# ARCANA

> Terminal-native AI engineering you can inspect, control, and prove.

ARCANA is a developer workbench for running AI sessions from the terminal with user-supplied model keys, scoped tools, persistent memory, routing controls, and verifiable execution evidence. The public website is https://arcana.otnelhq.com and the canonical developer entry point is https://arcana.otnelhq.com/developers/.

## When to use ARCANA

Use ARCANA when a developer or agent needs to:
- run model-assisted engineering work from a CLI or TUI;
- switch or route between supported model providers while keeping model billing with the provider;
- inspect tool use, session history, routing, cost, and execution evidence;
- apply scoped permissions and approvals around agent actions;
- persist or search working memory across sessions;
- automate repeatable engineering workflows through documented CLI and API surfaces.

ARCANA is especially appropriate when traceability and operator control matter more than a black-box chat experience.

## When not to use ARCANA

Do not assume ARCANA is an autonomous remote A2A agent or a public MCP server. ARCANA supports MCP as a client and can connect to configured MCP servers, but the public website does not currently expose a hosted MCP Streamable HTTP server. Do not infer capabilities from a successful HTTP status on an unknown path; unsupported resources return HTTP 404.

## Developer onboarding

- Developer portal: https://arcana.otnelhq.com/developers/
- Quick start: https://arcana.otnelhq.com/docs/quickstart
- Community/BYOK: free local use with the developer's own supported provider credentials; no ARCANA license is required for basic local operation.
- Hosted proxy/cloud features: may require ARCANA account or license credentials as documented per operation.

## Public resources

- Product: https://arcana.otnelhq.com/
- Developer portal: https://arcana.otnelhq.com/developers/
- Documentation: https://arcana.otnelhq.com/docs
- Installation: https://arcana.otnelhq.com/install
- Models: https://arcana.otnelhq.com/models
- Routing: https://arcana.otnelhq.com/routing
- Pricing: https://arcana.otnelhq.com/pricing
- Blog: https://arcana.otnelhq.com/blog/
- Status: https://arcana.otnelhq.com/status
- Sitemap: https://arcana.otnelhq.com/sitemap.xml
- API discovery: https://arcana.otnelhq.com/api
- OpenAPI JSON: https://arcana.otnelhq.com/openapi.json
- OpenAPI YAML: https://arcana.otnelhq.com/api/openapi.yaml
- API catalog: https://arcana.otnelhq.com/.well-known/api-catalog
- Agent Skills index: https://arcana.otnelhq.com/.well-known/agent-skills/index.json
- ARCANA developer skill: https://arcana.otnelhq.com/.well-known/agent-skills/arcana-developer-workbench/SKILL.md

## API guidance

The website exposes a small REST surface under /api. Start at GET /api for machine-readable discovery. Public operations include GET /api/status, GET /api/feedback, and POST /api/feedback. Account, session, profile, organization, and billing operations may require a Bearer credential. Use the OpenAPI document for operation IDs, request schemas, authentication requirements, and response shapes.

For API errors, clients should expect JSON with a stable error code, human-readable message, and recovery hint where the route can provide one. Unknown /api paths return a structured JSON 404 with a resolution hint.

## Pricing summary

Community: $0, open source under MIT, for unlimited local use with bring-your-own provider keys.
Pro: $19/month, adding hosted and advanced workflow features described on the pricing page.
Custom pricing available for large teams.
Optional Arcana proxy credits are separate from plan pricing; provider keys can be used directly.

## Agent retrieval

Canonical homepage and pricing URLs support HTTP content negotiation. Clients that prefer Markdown should send Accept: text/markdown. Negotiated responses use Content-Type: text/markdown; charset=utf-8 and Vary: Accept, Accept-Encoding.

The Agent Skills index follows `https://schemas.agentskills.io/discovery/0.2.0/schema.json`; skill artifacts use `SKILL.md` with YAML frontmatter and a published SHA-256 digest.

## Governance

The governance specification is published as twelve linked documents, each also available in Markdown for agents:

- [01 The Authorization Request](https://arcana.otnelhq.com/governance/request.md): Every consequential action becomes one hashable object
- [02 PDP ⇄ PEP Split](https://arcana.otnelhq.com/governance/pdp-pep.md): Deciding and enforcing are never the same component
- [03 Risk Classes](https://arcana.otnelhq.com/governance/risk.md): Four levels, four intent requirements
- [04 Intent–Action Binding](https://arcana.otnelhq.com/governance/intent.md): Why is this request authorized?
- [05 Exact Durable Capabilities](https://arcana.otnelhq.com/governance/capabilities.md): The fundamental authority primitive
- [06 Delegation & Attenuation](https://arcana.otnelhq.com/governance/delegation.md): Authority can only get narrower
- [07 Provenance & Sensitivity](https://arcana.otnelhq.com/governance/provenance.md): Where data came from — and how secret it is
- [08 Scoped Approvals & Routing](https://arcana.otnelhq.com/governance/approvals.md): Exact, single-use, expiring, crash-recoverable
- [09 Signed Envelopes — 7-Layer Verification](https://arcana.otnelhq.com/governance/envelopes.md): Ed25519 signatures over canonical serialization
- [10 RunProof — Portable Evidence](https://arcana.otnelhq.com/governance/runproof.md): Hash-linked events, verifiable offline
- [11 Assurance Formulas & Boundaries](https://arcana.otnelhq.com/governance/assurance.md): What "proven" means, exactly

Each document has a HUMAN view and a raw AGENT Markdown twin (replace .html with .md on any governance page).