[ DOCUMENTATION ]

Oddysey documentation

Everything about running an agent fleet that watches the market for you and never trades without you. Start with the quickstart if you want the whole loop working in ten minutes; start with the concepts page if you would rather understand it first.

Read in this order

The sidebar is the reading order. Each page ends with a link to the next one, so you can also just keep going.

START HERE

What Oddysey isA supervision layer for portfolio agents — and the one thing it deliberately refuses to do.QuickstartSign in, arm a watch, review a drafted proposal, and close the loop — in about ten minutes.

CORE CONCEPTS

Concepts and glossaryOperators, watches, proposals, the ledger, and the vocabulary the rest of the docs assume.The Command DeckA tour of every screen in the deck and what each one is for.

TUTORIALS

Arming watchesThresholds, actions, agents, cooldowns, and how to tune a watch that fires too often or never.Proposals and approvalHow a draft is built, how to read the allocation table, and what happens after you approve.Write your own agentCommission an agent from a brief, give it a schedule and a slice of the deck, and read what it reports back.Connect your own agentIssue a deck token, point an MCP client at Oddysey, and let your agent execute what you approved.

REFERENCE

Inference APIOddysey's OpenAI-compatible endpoint: one deck token, one balance, no model-provider account of your own.REST APIEvery endpoint the deck itself uses: envelope, auth, status codes, payloads.Self-hostingRun Oddysey locally and deploy your own instance to Cloudflare Workers with D1 and cron.Security modelSessions, token hashing, encryption at rest, and the blast radius of every credential.FAQ and troubleshootingPlaceholder prices, silent watches, 401s over MCP, and the other things that go wrong first.

The one-paragraph version

You arm a watch on a symbol. A scheduled sweep checks that symbol every fifteen minutes against your threshold. When it trips, an agent drafts a proposal — a title, a rationale, and a target allocation — and puts it in front of you as pending. You approve or reject it. Oddysey never places the order: your own agent reads the approved proposal over MCP, executes it on whatever rail you actually use, and reports the fill back so it lands in the ledger.

If you only read three pages

Quickstart — the entire loop, end to end, with a demo session so you do not need a wallet.
Connect your own agent — the part that makes Oddysey useful rather than a dashboard.
Security model — what each credential can do, and what happens if one leaks.