Aletheia is the retrieval-first knowledge substrate your AI talks to directly — over the Model Context Protocol. Voice recordings, meeting transcripts, scattered markdown — captured. Entities resolved, links drawn, contradictions surfaced — automatically. Nothing to file. One corpus, queryable from any LLM, agent, or app via MCP. Memory you can rely on, owned by you.
what's going on with the orion rollout? blockers, owners, what's next.
You record every client call with Plaud. You run Granola on every Zoom. Otter on the rest. Your inbox fills with transcripts. You re-read maybe one in twenty.
You ask an LLM to remember things, and sometimes it does. You can't see what it kept, can't audit when it was wrong, can't take it with you when the model changes.
You tried Notion. You tried Obsidian. Both made you the filing clerk for your own brain. You wanted thinking; they wanted hierarchy.
Aletheia is the substrate underneath. You capture. It extracts, links, resolves, and surfaces — automatically. You query from whichever LLM you're using today, against a knowledge layer you actually own.
Most "chat with your notes" tools are a vector search with a chatbot bolted on. Aletheia is structured retrieval first, generation second.
Purpose-built workflows hunt for decisions, blockers, ownership chains, and open questions — not just keyword matches. Ask a question, get a synthesis. Seven workflows, routed by intent.
Every claim links to the exact note, with timestamp and capture source. Inline pills you click to jump straight to source. No "according to my training data" — if we don't have a source, we say so.
Postgres + pgvector under the hood. Universal markdown plus YAML in, universal markdown plus YAML out. Every supported tool round-trips. No proprietary format you can't escape.
Auto-resolves people, companies, projects. Builds the entity graph as you capture. Co-occurrences, n-hop neighborhoods, relationship paths, entity timelines — without you filing anything.
Plaud, Granola, Otter, Limitless — every voice-capture tool ships its own thin summary feature and drops the transcript in your inbox. Aletheia is the substrate underneath them. You record the way you already record. The structure happens on the other side.
The transcript that used to die in your inbox becomes part of a queryable graph the same minute it arrives. Capture-and-forget.
Each workflow is a separate retrieval + reasoning chain tuned for a specific shape of question. You don't choose — Aletheia routes based on intent. Available in the built-in chat or to any LLM you connect via MCP.
Status, decisions, open questions, next actions for any topic.
Entities, tags, and supporting passages connected to any subject.
Chronological narrative of any topic, every beat linked to source.
Filtered, scoped search with rich previews and provenance.
Verbatim passages with context windows, ready to drop into a draft.
Open action items extracted across notes, grouped by owner.
Conflicting statements surfaced across notes with both sources side-by-side. The one workflow nothing else does.
Multiple clients, each with their own thread of context. Five recorded calls a week, sometimes ten. Currently using Plaud or Granola; transcripts pile up faster than synthesis.
Five live projects, dozens of meetings a week. Notes scattered across Joplin, Slack threads, email summaries. Status updates that should take ten minutes take ninety.
Discovery interviews, investor conversations, team check-ins. Captures everything because it's all signal — and re-reads almost none of it.
Long-horizon client engagements. Tracks themes, commitments, and shifts across months across many clients.
Reading and note-taking across hundreds of sources. Threading arguments, finding contradictions, building timelines. Citation discipline is the job.
Aletheia is overbuilt for some workflows and unsuitable for others. We'd rather you self-select than churn at week three.
Under five substantive captures a week, your LLM's built-in memory or a plain notes app is enough. Aletheia is overbuilt for you.
If local-first is your hill, we're not it. We respect the philosophy and won't try to talk you out of it.
Notion is a great canvas. Aletheia sits underneath it, not next to it. Try us if you want a substrate; not if you want a replacement.
That's not what this is. There are good tools for that question — we're not one of them.
The category ships the same three features in a trenchcoat — chat-with-your-notes, auto-tag, summarize-this-page — and stops there. Aletheia is a different bar. Here's the honest map of what we replace, complement, and ignore.
A model's memory is opaque, lossy, and vendor-held. You can't audit it, query it directly, or take it with you when the vendor changes the format. Aletheia is the substrate underneath; the model is one of several reasoning heads on top. Mature setups have both.
Notion is a collaborative document canvas with a chat sidebar bolted on. Aletheia is the entity graph and temporal layer over your captures. Different jobs. Use both.
They ship vector search and a chat box. Aletheia ships entity resolution, temporal reasoning, and contradiction detection. Different depth. Their retrieval flatlines as the corpus grows; ours improves.
Obsidian is brilliant if you love filing. Aletheia is for people who don't have time. We import from Obsidian; many users keep both.
Tana asks you to design supertags and queries before the value shows up. Aletheia asks you to capture and lets structure emerge. Same destination, opposite architectural bet.
Mem, Saga, Reflect, and a dozen others ship the same three features bolted onto a notes UI with no opinion on retrieval quality or graph reasoning. Aletheia isn't an AI notes app. It's the substrate underneath.
Hybrid retrieval (BM25 + embeddings + tag-match), section-boosted rerank, then the right synthesis workflow for the question. Exposed via MCP so any model can call it.
structured chunks (## decisions, ## open questions) get amplified.
every claim → click → source note.
switch the client, not the substrate.
No lock-in. Universal markdown is the substrate — every supported tool round-trips.
One-click import. Drag the export, watch the count tick up, ask a question.
Markdown plus YAML frontmatter, packaged as a ZIP. Opens in any tool, forever.
$ aletheia export --format=md --out=./vault.zip
Flat-rate pricing for managed infrastructure, synthesis pipelines, and MCP access. No surprises, no over-the-line bills.
Tap to expand. We tried to answer the questions you'd actually have, not the ones marketing wants you to ask.
Create a workspace, drop in your vault, ask the first real question of your own corpus. It takes about four minutes.