Getting started
Your first document base
Ten minutes from a folder of files to answers with sources. You name a document base, add documents, pick a model, and let onboarding compile them into resident memory. Everything after that is asking questions.
Before you start
Engram is invite-only in production today. The sign-in page at app.engramdynamics.org carries a Need access? Request an invite link, and that is the way in: ask for an invite, or use the invite link you were sent. Nothing on this page works until you can sign in.
If your workspace signed up rather than being invited, confirm your email address first. You can sign in and look around unverified, but the moment you try to add documents the app says Confirm your email address to start adding documents and points at the link in your inbox. One click and you are through.
Buying a plan is a separate step and has its own gate: the Service Agreement and the DPA have to be accepted in Settings > Legal before checkout will run. See terms and DPA acceptance.
Create the document base
From the dashboard, choose New Document Base. The wizard has five steps and it is resumable: leave at any point and the dashboard brings you back to the step you were on.
- Name. What you want the AI to know, named for humans. "Support KB", "Sales collateral", "2026 policies". Up to 200 characters.
- Documents. Drag files or a folder in. Details below.
- Model. Pick the model tier the base is compiled for. Tiers that are not live yet render as coming soon.
- Review. Document count, model, and the estimate, in one screen before anything is built.
- Onboard. The build itself, with live progress.
Names are how you refer to a base everywhere else, including the CLI and MCP, so pick one you will recognise in a config file a month from now.
Add your documents
The Documents step accepts .txt, .md, .pdf, .docx, .doc, .html, .xlsx, .xls and .csv. Anything else is dropped from the drop with the reason shown, so a folder with images and archives in it still works: only the readable files go up. Empty files are skipped too, because a blank document is only ever a bad answer waiting to happen.
A document is a unit of text, not a file. In the platform's own words:
A document is up to 4,000 tokens of extracted text, roughly six pages. A longer file counts as several, so a sixty-page report counts as ten documents. Adding files is always free. You pay to keep them ready to answer from, not to load them.
Your plan's document, query and workspace allowances live on plans and allowances. Concepts and limits explains what the platform counts and when.
Watch onboarding finish
Onboarding reads each document once and compiles it into a cartridge that lives in resident memory on the serving hardware. It runs as a job you can watch: the wizard shows live progress, and the document base's page keeps the same view afterwards.
Each document walks its own lifecycle, and the two that matter to you are ready (answerable) and failed (say so and it can be re-sent). A document base whose documents are all ready has status ready. A base that is still building is training. Full state list on Concepts and limits.
Onboarding is not something you have to sit and watch. Close the tab, come back later, or check from the terminal:
engram status "Support KB"
Ask your first question
Open the document base and use the chat tab. Ask something a document actually answers, and the reply names the documents it drew on. That is the same answer path everything else uses: the REST API, the MCP tools your agents call, and the CLI.
Two things make answers better straight away. Ask self-contained questions, and raise the number of documents consulted when a question spans the whole base rather than one file. Over MCP those are the question and k arguments of query_corpus.
The same thing from the terminal
If you would rather not click, the CLI does the whole flow, and it is the right tool for a folder bigger than a handful of files. Install it, create a key on the API keys page in the app, and push:
pipx install engram-dynamics
engram login --api-key <your key> --api-url https://api.engramdynamics.org
engram corpora create --name "Support KB"
engram push ./docs --corpus "Support KB"
engram status "Support KB"
engram push hashes locally, asks the server what it does not already have, and sends only that. Re-running it after a change is cheap, which is what makes it a good fit for a folder you keep editing. See Push a folder.