Your organization remembers — with dates and sources.

Your knowledge is scattered across emails, chats, wikis and tickets. Retivum pulls it into one graph of facts, so questions in the chat get answers with cited sources instead of guesses.

Start trial Book a demo

MCP tools on one endpoint

content sources in one ingest call

time axes on every stored fact

database behind graph and operations

Why it matters

Without memory

Every question starts from scratch. Someone re-explains the vendor decision every quarter, and nobody is sure which version of the policy counts today.

With memory

The answer, its date and its source are already there. One question returns the recorded decision, when it applied, and where it came from.

What this looks like on a workday

New hire onboarding

A new engineer asks why you chose this database. They get the recorded decision with its reasoning, the date it was made, and the ticket it came from.

Handover across teams

Support needs to know what was agreed with a customer last spring. Chat and email facts come back with validity windows, so what changed since is visible.

Audit and compliance

A compliance officer asks which policy was in effect on a given date. An as-of query returns exactly that snapshot, traceable to its source episode.

What goes in

Emails and letters — IMAP, Gmail and Outlook connectors
Chat transcripts from Slack and Microsoft Teams
Wiki and documentation pages from Confluence, Notion, SharePoint and Jira
Markdown, HTML, plain text and saved URLs
JSON and CSV tables, plus structured fact payloads
Images and office documents (PDF, DOCX, PPTX, XLSX)
PostgreSQL table rows — one saga per table
GitHub and GitLab issues, RSS feeds and Google Drive files

Issue trackers, chat and file platforms arrive through thin connectors that feed one ingest boundary over REST and MCP.

Everything you ingest becomes available to the AI through governed retrieval: full-text, vector and graph-traversal candidates merge into ranked answers with cited provenance and point-in-time as-of queries. Relationship types stay governed per group, and graph analytics — centrality, PageRank, shortest paths — run over exactly the groups a caller may see.

Who sees what

Knowledge is scoped per tenant and per group; searches cover only the groups a caller may see
Groups are private by default — visible to their owner and to the groups or people they are explicitly shared with
Roles govern the view: system_admin, tenant_admin, tenant_writer, tenant_reader and auditor
Every call is audited; the tamper-evident audit chain verifies per tenant through one endpoint

REST API

Versioned JSON API under /api/v1 with bearer API keys, tenant-scoped paths, pagination defaults, a single error envelope and health plus readiness probes.

MCP server

One Streamable HTTP endpoint for MCP clients, plus stdio for local use. Tenant and group presets keep client configuration short.

Command-line client

JSON-output CLI covering ingest, search, lookups, deletes, jobs, ontology and transfer — the same surface as the API, scriptable in CI.

Web inspector

Admin UI for the graph explorer, query console, ontology review queue, decisions, audit chain and jobs. Presentation only; the backend owns authentication.

Next: Data Protection →

Why it matters

Without memory

Every question starts from scratch. Someone re-explains the vendor decision every quarter, and nobody is sure which version of the policy counts today.

With memory

The answer, its date and its source are already there. One question returns the recorded decision, when it applied, and where it came from.

What this looks like on a workday

New hire onboarding

A new engineer asks why you chose this database. They get the recorded decision with its reasoning, the date it was made, and the ticket it came from.

Handover across teams

Support needs to know what was agreed with a customer last spring. Chat and email facts come back with validity windows, so what changed since is visible.

Audit and compliance

A compliance officer asks which policy was in effect on a given date. An as-of query returns exactly that snapshot, traceable to its source episode.

What goes in

Emails and letters — IMAP, Gmail and Outlook connectors

Chat transcripts from Slack and Microsoft Teams

Wiki and documentation pages from Confluence, Notion, SharePoint and Jira

Markdown, HTML, plain text and saved URLs

JSON and CSV tables, plus structured fact payloads

Images and office documents (PDF, DOCX, PPTX, XLSX)

PostgreSQL table rows — one saga per table

GitHub and GitLab issues, RSS feeds and Google Drive files

Issue trackers, chat and file platforms arrive through thin connectors that feed one ingest boundary over REST and MCP.

Who sees what

Knowledge is scoped per tenant and per group; searches cover only the groups a caller may see

Groups are private by default — visible to their owner and to the groups or people they are explicitly shared with

Roles govern the view: system_admin, tenant_admin, tenant_writer, tenant_reader and auditor

Every call is audited; the tamper-evident audit chain verifies per tenant through one endpoint