The unified context layer
for every AI you use
Put your durable knowledge in one place. Orcha indexes it into one searchable substrate and serves the right passages to any AI tool you authorize, over MCP or REST, with a citation on every result.
How it works
Put your knowledge in one place
Write natively, import an Obsidian vault or Notion workspace with its structure intact, model what you track as structured tables, and connect read-only sources. All of it becomes searchable the same way.
Indexed the way documents are written
Documents break at headings and section boundaries, and every passage keeps its heading path and line position, so results stand on their own and point back to their exact location.
Two kinds of search, one ranking
Keyword search and semantic search run over the same indexed passages and merge into one ordering, so it works whether someone remembers the exact phrase or only the idea.
Every agent, same rules
One MCP endpoint and a REST API expose the same context to any client you authorize, with permissions checked on every request and a citation on every result.
A home for the knowledge you decided is worth keeping
Whatever you put in, from a note to a table to a connected folder, becomes searchable the same way, so one question can pull from all of it.
Files and folders
The familiar place to keep what you know. Write in Markdown, organize with folders and tags, and import and export freely.
Structured databases
Tables for the things you track, like clients, projects, and decisions. Typed records are indexed and cited alongside your prose, and rows can link to the files that define them.
Obsidian and Notion migration
Your vault or workspace arrives intact: folders, tags, links, backlinks, and attachments, with a dry run showing exactly what will come over before anything changes.
Connected sources
Keep GitHub, Notion, Google Drive, Dropbox, Linear, and more in the picture without copying anything. Sources are read-only, and Orcha never writes back to them.
Citations on every result
Every answer points back to where it came from, with a deep link and freshness information. Check the answer in one click.
Access you control
Decide exactly what each AI tool may read, write, or delete. Scope access to folders or workspaces, set expiry, and revoke it at any time.
What ships today
Context management
- Markdown-native files with folders and tags
- Structured databases with typed records and relations
- Context bundles: named sets for a job, with routing guidance for agents
- Version history with restore and side-by-side diff
- Bulk import and export
- Obsidian and Notion migration with dry run
Retrieval
- Semantic and keyword search merged into one ranking
- Results as passages, not whole files
- Heading paths and line anchors on every result
- A citation on every result
- One query spans files, records, and connected sources
- Retrieval quality gated by benchmarks in CI
Access and delivery
- MCP server with 21 tools, connected over OAuth
- Verified connections for Claude and ChatGPT
- Token-authenticated REST API across every resource
- Read, write, and delete permissions per token
- Folder- and workspace-scoped access
- Controlled sharing links and signed webhooks
Teams and governance
- Organizations and workspaces with role-based access
- Admin, editor, and viewer roles
- Team invitations via email
- Permissions re-checked against live roles on every request
- Audit and activity logs
- Usage analytics: see which context your AI actually uses
For engineers
Files, structured records, and connected-source documents normalize into one indexed document and chunk model. Each chunk carries a 512-dimension vector and a Postgres tsvector, so keyword and semantic retrieval rank identical units and fuse with weighted RRF. Workspace scope, folder scope, and soft-delete exclusion are conditions in the retrieval SQL. Results come back as passages with heading paths, line anchors, citations, and an optional explain trace. Reach it through 21 MCP tools over authenticated Streamable HTTP or a token-authenticated REST API.
Read-only by design
Orcha never writes back to your connected sources.
Cited by default
Every result carries a link to where it came from.
Yours to leave with
Full export of files and structured data, any time.
Set your context up once
Orcha is not open to everyone yet. Join the waitlist and we will let you know when a spot opens up.