MESSAGE AInitial plan
“The workshop will be hybrid, with a group at the venue.”
HELP AI NAVIGATE LARGE INFORMATION SPACES01 / MAILWEAVE
Help your AI
find what matters.
Orivra helps AI agents navigate large information spaces efficiently: find the right information, understand what happened and why, and bring back evidence you can check.
0.2.0-beta.1 · Apple Silicon · macOS 14+ · 346 MB
Gmail access is unverified. Selected email goes to Claude.
Installation & requirements · Release notes
A LARGE INFORMATION SPACE
A conversationA documentAn updateRelevant evidenceWhere to look next
A PATH TO WHAT MATTERSCONCEPT VIEW
Find what matters.Understand what happened.See what supports it.
Your agent has access.
Can it find the answer?
As information grows across conversations, files and tools, finding the right piece becomes a task of its own. An agent has to decide where to look, what to read and when it has enough to answer.
Orivra gives it a way through: find relevant evidence, explore what it needs next without loading everything at once, and keep a path back to the source. So your agent can help you understand what happened and why—not just return a list of matches.
01 / THE APPROACHFind it. Follow it.
Find it. Follow it.
Understand it.
Bring the useful pieces into view.
Read deeper where the question leads.
Keep the evidence with the answer.
Get to the information the question needs.
Search for relevant evidence and bring useful pieces into view. Your agent can start with those pieces instead of reading every conversation in full.
ILLUSTRATIVE EXAMPLENOT A LIVE SEARCH
“Is the workshop still in person?”
Read the follow-up
MESSAGE BUpdated plan
“The venue lost step-free access. The final format is remote-only. This replaces the hybrid plan.”
Two relevant messages · one conversation
Fictional messages illustrating the workflow—not a measured result or a product screenshot.
UNDER THE HOOD
Underneath the navigation is a query-time context graph: a map of the evidence retrieved for a question and the relationships between it. Your agent starts with a compact view and can ask for more detail where it needs it. The map is built for the question, not for your entire workspace.
02 / THE VISIONMore places to look.
More places to look.
One way to explore.
The vision reaches beyond any one source.
Beginning with the inbox.
Video description
03 / EVIDENCE, NOT ASSUMPTIONAn answer should
An answer should
leave a trail.
Finding information is only useful if you can see what supports the answer—and what is still missing.
Trace it back.
Keep source references with the evidence, so you can inspect the source behind a claim.
Separate fact from inference.
See what a source actually says, what changed, and which connections are inferred rather than explicitly stated.
Make gaps visible.
Show what was left out and how to read more when possible. “Not enough evidence” remains a valid answer.
Keep the sources, relationships and limits in view—so you can judge the answer, not just accept it.
04 / MEET MAILWEAVEPUBLIC EXPERIMENTAL BETA
Start with
your inbox.
MailWeave is Orivra’s Gmail engine. It helps Claude find the email that matters, read beyond a search preview, and follow a conversation to the messages that explain the decision.
Get started with MailWeaveMIT-licensed source. Early release with known limitations; repository CI and release checks are not fully green.
First connectorMailWeave for Gmail
First clientClaude Desktop
Mailbox permissionRead only. No sending or editing.
ProcessingLocal retrieval and local ranking models
What Claude receivesSelected email content and references
Next sourcesSlack + Google Drive Planned, not included
A FEW GOOD QUESTIONSBefore you
Before you
connect.
Does Orivra replace Claude?
No. Orivra helps your agent search and navigate information through tools connected over MCP. Claude uses those tools, reasons over the evidence and writes the answer. You keep working with your existing AI assistant.
Is this only for Gmail?
No. Orivra is built around helping AI agents navigate large information spaces. MailWeave is its first engine, for Gmail. Slack and Google Drive are planned; they are not included in this preview.
Does it guarantee a correct answer?
No. Orivra gives your agent ways to find and inspect evidence, but searches can be incomplete and an AI can misinterpret what it reads. References and visible gaps help you check its work. Comparative accuracy and efficiency benefits are still being evaluated.
Does it build a graph of my entire mailbox?
No. It builds a bounded context graph for a question, using the evidence retrieved for that request. It is not a permanent, complete map of your mailbox, and it does not claim to have read everything.
Does local-first mean my email stays off the cloud?
No. Retrieval and local model processing run on your computer, but selected email content is returned to your connected AI client. In Claude Desktop, that content can be sent to Anthropic and retained under your Claude settings and its policies.
Can it send or change my email?
No. The Gmail connector requests gmail.readonly. It cannot send, delete, archive or label messages. You can revoke its access in your Google Account.
Can I download it today?
Yes. MailWeave 0.2.0-beta.1 is a public experimental beta for Claude Desktop on Apple Silicon Macs running macOS 14 or later. Download the .mcpb installer here, then follow Getting started. The source is public under MIT, and the advanced, self-managed route remains available.
Has Google verified the Gmail connection?
Gmail data-access verification is not complete. Google may show an unverified-app warning, enforce a 100-total-user cap, or block authorization under account or administrator policies. The cap is not a verification exemption. This is an experimental release, not a Google-approved app.
Give your agent
a way through.
Start with MailWeave for Gmail, or tell me where your agent gets lost. Orivra’s direction is bigger: helping agents navigate the information you rely on, wherever it lives.
Talk to the builder Direct email to Nayan. No signup form or mailing list.