A state layer for agentic workflows

State your agents share.
Today, tomorrow, and the agent you install next week.

Point your coding agent at it, drop in a folder of PDFs, or wire up your custom one. Ouroboros parses, structures, and remembers — so the next session, and the next agent, and the agent you switch to in six months, all start where the last one left off.

Private beta · BYOK (Claude Code, Codex, Cursor, your own) · Local-first

The thesis

Pay once. Query forever.

Most teams don't notice the pattern. Every fresh agent session re-reads the same documents from scratch. Ask your agent about a 200-page lease today — 60,000 tokens, ten minutes, one answer. Ask again tomorrow in a new session — another 60,000 tokens, another ten minutes, the same answer. A week of that across a few agents and a handful of documents is half a million tokens of pure rework. Your agent is slow the whole time, waiting for the same parse to complete before producing the same answer it produced yesterday.

Same math for code. A coding agent on a fresh repo burns context before it writes a single line — globbing the file tree, reading a dozen files to map the architecture, following imports, hunting for callers. Every new session repeats it. Every agent you switch to repeats it. The repo didn't change; the cost did.

Without Ouroboros

Pay every time.

  • First read: 60k tokens · ~10 min
  • Same question tomorrow: 60k tokens · ~10 min
  • 10 questions across 5 agents: ~3M tokens of rework
  • Cost trajectory: linear in queries — every agent re-derives every answer.

With Ouroboros

Pay once. Query forever.

  • One-time extraction: a few thousand tokens, the document lands in the graph.
  • Same question tomorrow: a few-thousand-token SDK query, sub-second round-trip.
  • 10 questions across 5 agents: the one-time extract, plus ~50k tokens of cheap queries.
  • Cost trajectory: flat after extract — every agent gets the same indexed answer.

The compounding is the whole point. Documents you ingested last week — or modules you parsed last week — aren't dead weight in your token budget. They're free knowledge for every agent you'll ever connect, today and a year from now.

The system

One daemon. Three layers.

Files on disk you can open without it. A typed knowledge graph your agents read through one connection. A gate between them that nothing crosses without your permission.

The gate

Every agent connection is minted by a click in the tray app — never by a config file, never by an HTTP call. Each connection has an explicit scope: which entities, which actions. Read it back, change it, or revoke it from the same tray.

The engine

Drop a folder, get a typed knowledge graph. Drop a repo, get a symbol graph with every import, call, and reference resolved. Every claim mined from a document carries a verbatim quote that's substring-checked against the source — claims whose quotes don't literally appear get dropped. The check is a for-loop, not a model.

The vault

Documents are content-addressed at ~/Ouroboros/vault/sha256/. Wiki pages are plain markdown at ~/Ouroboros/vault/wiki/. Open them in Obsidian; the wikilinks, embeds, and tags work natively. The encrypted SQLite is a projection of the files, not the other way around.

architecture: gate → engine → vault
flowchart LR
    A1[Claude Code] --> G
    A2[Codex] --> G
    A3[Cursor] --> G
    A4[Your agent] --> G
    G[The gate<br/>per-connection scope<br/>session-key auth<br/>audit log every read]
    G --> E[The engine<br/>tree-sitter ingest<br/>quote-verified extraction<br/>hybrid retrieval]
    E --> V[The vault<br/>content-addressed files<br/>plain-markdown wiki<br/>encrypted libsql]
    V -.opens directly in.-> O[Obsidian / your editor]

Explore deeper

Each piece, in detail.

The system is a small set of primitives that compose. Each one is its own page — what it does, how to use it, what it earns you, where it's headed.

Trust Covenant

Five rules. Each one is enforced by code, not by a promise. The 1Password / Signal trust model applied to agentic state.

MCP & the Sophia SDK

How agents actually talk to Ouroboros — Model Context Protocol, the TypeScript SDK, and the V8 isolate that turns N round trips into one.

Agentic Coordination

When two or more agents share a project, they shouldn't have to discover each other through you. Channels, inbox, identity attestation, and the side-channel that makes 'you cannot miss messages' a hard guarantee.

Data Structuring

How a folder of PDFs becomes a queryable knowledge surface — three honest ingest tiers, profile-aware mining, hybrid retrieval, and quote-grounded extraction the model cannot fake.

Knowledge Graph

The structured-data thesis. Why typed entity-predicate-object rows with dated observers and grounded quotes beat in-context everything — and where we're heading past the Karpathy-wiki ceiling.

Codebase Graph

Point Ouroboros at a repo. Tree-sitter parses every module locally — modules, symbols, edges — into a typed graph your coding agent queries in sub-second time. The orientation toll your agent pays on every fresh session, paid once.

Wiki Primitive

Plain-markdown wiki pages on disk. Open them in Obsidian. Sign them with ed25519. Scope reads to entities. The wiki is your durable scratchpad — and the agents' too.

Time Machine

Every write is journaled with full before/after JSON. Any row, any time, revertable. System actors can never hard-delete user data — only soft-deactivate, tombstone, or supersede.

What's running today

One real instance. One real life's worth of data.

Ouroboros is in private beta. One person uses it daily — the person building it. No paying customers, no waitlist soup, no "trusted by thousands of researchers." Every number on this page is from one real instance, ingesting one real life's worth of documents and code.

4,928
documents indexed
15,202
entity-attributed facts
every one with a verbatim quote
2,998
code modules
across 6 languages
13,554
symbols resolved
8,407
edges in the code graph
120
agent tools
11
entities

Live state from May 2, 2026, the machine that built this.

Get it

You can't, yet.

Ouroboros is in private beta and the source is closed. We're not taking sign-ups, running a waitlist, or selling early access. The page you're reading is the marketing surface; the product behind it is still the dogfood instance.

When the source opens and packaging ships, the install will be one command and you'll be able to point your own agents at your own machine. Until then, the right thing to do is wait. There is nothing to buy and nothing to download.

If you're an MCP-curious developer, a researcher with a folder of PDFs, or a privacy-conscious power user — keep reading the deep-dive pages so you'll recognize what's shipping when it does.