Auditable, editable memory for AI applications

Odyssey is memory infrastructure for AI systems. Ingest knowledge, inspect what was retrieved and why, edit memory structure directly, and govern access across projects.

Memory Inspector
acme-copilot
Query: "How does authentication work?"
#1

Authentication flow uses OAuth 2.0 with PKCE for secure token exchange...

auth-spec.md0.94
#2

Token refresh happens automatically when the access token expires...

session-handler.ts0.87
#3

Rate limiting is enforced at 100 requests per minute per API key...

api-limits.md0.81
3 chunks retrievedlatency: 124ms

What Odyssey is

Odyssey is memory infrastructure for AI applications. It combines memory ingest, retrieval inspection, memory graph editing, project scoping, and usage enforcement into a single operational layer.

Memory ingest

Push documents and chunks via API

Retrieval inspection

See what was returned and why

Graph editing

Edit entities, relations, and tags

Project scoping

Isolate memory per project

Usage enforcement

Set limits and track consumption

What you can do with Odyssey

Inspect retrieval

See exactly which chunks were retrieved for every query. View scores, ranking, source documents, and token counts. Understand why the system returned what it did.

Edit memory structure

Manipulate memory directly through entities, relations, and tags. Link chunks to graph items. Adjust how memory is organized without re-ingesting.

Scope memory per project

Every document, chunk, entity, and relation belongs to a project. Retrieval is isolated by default. Build multi-tenant AI products without cross-contamination.

Govern usage and access

Set quotas per project and API key. Track ingest and retrieve calls. Enforce limits before costs spiral. Control who can access what memory.

How it works

Odyssey is both a web application and an API. Use whichever surface fits your workflow.

01

Create a project

Set up a project to scope all memory, keys, and usage. Each project is a separate memory domain.

02

Ingest memory

Push documents, chunks, or structured data via API. Tag with metadata and link to entities.

03

Retrieve and inspect

Query memory and view full retrieval traces. See scores, sources, and ranking decisions.

04

Edit and govern

Manipulate entities, relations, and tags in the Studio. Set usage limits in the Dev Console.

See why the system returned this

You are not guessing why the AI returned something. You can inspect it directly: query, matched chunks, relevance scores, source documents, and linked entities.

Retrieval Trace
trace_8f2a9c
billing-service2.3s
Query
"How does metered billing work?"
Retrieved chunks (3)
#1

The billing module calculates usage based on metered events. Each event is tagged with a customer_id and project_id for attribution.

0.96
billing-architecture.md
BillingModuleMeteredEvent
billingcore
47 tok
#2

Metered billing events are batched every 60 seconds and written to the events table with deduplication by event_id.

0.89
event-pipeline.ts
EventPipeline
billingevents
38 tok
#3

Usage summaries are computed nightly and stored in the usage_daily table partitioned by customer.

0.84
usage-aggregation.sql
UsageSummary
billingaggregation
29 tok
Total tokens: 114Chunks scanned: 847

Web app and API

Odyssey is equal parts web application and REST API. Use the dashboard for exploration and debugging. Use the API for production integrations.

Web Application
Memory Explorer
Browse documents, chunks, and retrieval logs
Studio
Create and edit entities, relations, and tags
Dev Console
Inspect retrieval traces and debug responses
Projects
Manage project settings, keys, and team access
REST API
Ingest and retrieve
# Ingest memory
POST /api/v1/ingest
{
  "project_id": "acme-copilot",
  "chunks": [{ "content": "...", "source": "..." }]
}

# Retrieve context
POST /api/v1/retrieve
{
  "project_id": "acme-copilot",
  "query": "How does auth work?",
  "top_k": 5
}
Project-scoped memory access. Full retrieval traces. Usage tracked per key.

Start building with auditable memory

Odyssey is free to start. Set up project-scoped memory with full retrieval inspection in minutes.