Odyssey is a governed memory layer for AI systems. Ingest knowledge, retrieve grounded context, enforce usage, and inspect the evidence behind every retrieval across projects, agents, and customers.
Retrieved chunks, source documents, and references stay visible and inspectable.
Requests are tied to a project and API key, with access boundaries enforced server-side.
Retrievals, latency, and downstream model calls can be inspected after the fact.
Usage caps protect you when a tenant or agent starts going sideways.
{
"items": [...],
"count": 5,
"meta": {
"project_id": "proj_...",
"graph_used": true,
"api_key_id": "key_..."
}
}A vector database can store embeddings. It does not automatically give you tenant isolation, evidence visibility, request auditability, or usage enforcement. That gap is where agent systems get expensive, unstable, and hard to trust.
Without hard project boundaries, your retrieval layer becomes one shared soup of documents, embeddings, and accidental cross-tenant risk.
When an answer is wrong, most teams cannot quickly see which chunks influenced it, which query path was used, or how the system got there.
Agents can retrieve forever, spend unpredictably, and hammer infrastructure unless limits and counters are enforced where requests actually execute.
It sits between your application and your retrieval layer. You keep your product UX. Odyssey governs how knowledge is ingested, scoped, retrieved, measured, and audited.
Documents, chunks, API keys, logs, and usage all stay bounded to a project so each customer or environment behaves like its own memory domain.
Every retrieval returns traceable memory items and can be inspected later. You can see what influenced the result instead of guessing.
Ingest and retrieve calls are checked against limits per project and API key so cost and abuse controls live in the execution path, not in wishful middleware.
Odyssey is designed to be your memory infrastructure: state is scoped, access is enforced, requests are measured, and outputs stay tied to evidence. That is what makes memory survivable in production.
Treat each customer, workspace, or environment as its own governed memory domain.
Requests run under explicit key scope and project scope, not vague client-side assumptions.
Track retrievals, model calls, latency, and usage as first-class signals.
See what was retrieved, why it ranked, and what influenced the result.
Upload source material like SOPs, specs, tickets, contracts, and internal docs. Odyssey chunks, embeds, and stores them inside a project boundary.
Your app or agent calls Odyssey instead of raw vector search. Retrieval blends lexical, semantic, recency, and graph-aware signals.
Review evidence, request logs, API key activity, and usage counters so you can debug behavior and clamp cost before customers do it for you.
Odyssey gives you the operational surface around memory, not just a place to throw vectors.
See the current project, memory footprint, and the core flow through ingest, retrieval, and governed usage.
Inspect documents, chunks, entities, and relations instead of treating your retrieval layer like a black box.
Bring in raw knowledge and turn it into retrieval-ready memory with project scoping baked in.
Inspect request activity, API key usage, endpoint behavior, and operational signals in one place.
The interface stays simple. The behavior underneath does the heavy lifting: project scoping, usage checks, evidence logging, and retrieval control.
curl -X POST "https://your-domain.com/api/v1/retrieve" \
-H "x-api-key: $ODYSSEY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"query": "What changed in the vendor renewal policy?",
"limit": 10
}'Ground responses on project-scoped knowledge while preserving evidence and usage visibility across customer workspaces.
Replace fragile custom memory glue with a governed retrieval layer that can be measured, replayed, and constrained.
Use logs, limits, and request-level visibility to catch regressions, debug bad answers, and enforce cost boundaries.
If your AI product needs memory that can survive real customers, real traffic, and real cost, Odyssey gives you the missing layer: scoped memory, evidence-first retrieval, and hard operational controls.