Point an agent at a folder.
Get a project.
Meeteor now has a command-line tool. It writes to the same database as the app, through the same pipeline, so anything that can run a command on your PC can build a project for you: Claude Code, Codex, Cursor, a script, or you in a terminal.
From a messy folder to a project that answers, in one prompt.
You already have the material: contracts, decks, spreadsheets, email threads, meeting notes, a repository. The slow part was reading it all and typing the brief, the notes and the glossary. An agent does that part now.
- 01 · point
Name the folder
Tell your agent where the files are and what the project is for. It reads them the way it reads a codebase.
- 02 · write
It writes the context
A brief, a pinned status note, open questions, and a glossary of people, companies, products, acronyms and numbers.
- 03 · import
It runs meeteor-cli
Backup, create, brief, ingest, notes, glossary. Every file is parsed, chunked and indexed exactly as if you had dropped it on the app.
- 04 · call
Open Meeteor and take the call
The project is active. Related snippets, cards and Ask answer from what the agent loaded, with the source on screen.
Say what the project is for. The agent does the reading.
This is the whole instruction. The agent reads the files, decides what matters, writes the brief and notes in its own words, and calls the CLI for each step. Notes it writes carry the date, so the next time you refresh the project you know what was true when.
- Back up first. One command, one zip, before anything is written.
- Descriptive file names. A good agent copies Q3-pricing-final-v3.xlsx in as Acme pricing proposal, final (Sep 2026).xlsx, so citations read well on a card.
- Notes with a date in the title. “Where things stand (Oct 26)” beats “Status” when you refresh next month.
Read everything in C:\Clients\Acme (the MSA, the order form, the last three proposals and pricing.xlsx) and the thread in acme-renewal.eml. Then build a Meeteor project with meeteor-cli: 1. back up Meeteor first 2. create a project called "Acme Renewal" and write a brief: what the deal is, who is on each side, what we want, what is open 3. ingest the contracts, proposals and the pricing sheet 4. add a pinned note "Where things stand" and a note "Questions to ask on the call", both dated today 5. import a glossary of the people, companies, products, acronyms and the numbers that matter (cap, notice period, seats) 6. run a search for "uplift cap" and show me what comes back
Eight commands. The same pipeline as the app.
Each command goes through the functions the app itself uses. Documents are parsed on your PC (PDF with OCR, Word, PowerPoint, Excel, CSV, Markdown, HTML, email, images), split into passages with their page, slide or row, and indexed for both keyword and meaning search. Notes and glossary terms get the same treatment, so a seeded project retrieves exactly like one you built by hand.
- ingest skips a file the project already has (same content, same document), so re-running is safe
- glossary import merges by term name: run it again with more terms and nothing is duplicated
- note add adds a note every time; replace a note from the app when it goes stale
- search and ask let the agent check its own work before you take the call
meeteor-cli backup meeteor-cli project create --name "Acme Renewal" --activate meeteor-cli project brief --project "Acme Renewal" --file brief.md --description "Renewal of the Acme MSA, due November 30" meeteor-cli ingest --project "Acme Renewal" msa.pdf order-form.pdf proposal-v3.docx pricing.xlsx acme-renewal.eml meeteor-cli note add --project "Acme Renewal" --title "Where things stand (Oct 26)" --file status.md --pin meeteor-cli note add --project "Acme Renewal" --title "Questions to ask on the call" --file questions.md meeteor-cli glossary import --project "Acme Renewal" --file terms.json meeteor-cli search --project "Acme Renewal" "what is the uplift cap"
Names, acronyms and numbers, in a file.
A glossary is a JSON array. Seven kinds: person, org, product, acronym, term, number and place. Aliases matter: “Dana” on the call should find “Dana Whitfield” in the project. Terms also bias transcription, so the names your industry uses come out spelled right.
- A definition of a sentence or two becomes a searchable passage of its own
- A role (“CFO at Acme, signs the renewal”) tells the cards who is speaking
- Imported terms are accepted terms; the app's glossary editor can change or delete them
[
{ "term": "Acme Corp", "kind": "org", "aliases": ["Acme"],
"definition": "The customer. MSA signed 2024, renewal due November 30." },
{ "term": "Dana Whitfield", "kind": "person", "aliases": ["Dana"],
"role": "CFO at Acme, signs the renewal" },
{ "term": "MSA", "kind": "acronym",
"definition": "Master services agreement; section 7.2 caps the annual uplift" },
{ "term": "uplift cap", "kind": "number",
"definition": "5% per contract year (MSA 7.2)" }
]A run you can read, then a project you can open.
The CLI prints one line per document, note and term, with the chunk counts, and “ready” or the error for each file. When it finishes, open Meeteor: the project is there with every source, and the search the agent ran at the end is the same search the overlay runs during a call.
- Works with any agent or shell: Claude Code, Codex, Cursor, PowerShell, a scheduled task
- Runs against your Meeteor data folder; close the app while it writes
- Nothing leaves your PC: parsing, OCR and indexing are local and need no AI key
What meeteor-cli can do today.
The commands an agent uses to build and check a project. Evaluation and benchmark commands exist too; they are for developing Meeteor and are documented in the repository.
| Command | What it does |
|---|---|
| backup | Zips the whole data folder (database, originals, audio) to one file. Run it first. |
| project create · activate · list | Creates a project by name, makes it the active one, or lists them with their ids. |
| project brief | Sets the brief from a Markdown file and an optional one-line description. The brief is read by every prompt and by the transcription bias list. |
| ingest | Queues files into the project through the app's pipeline and waits until each is ready, printing the chunk count or the error. A file already in the project is reported as existing, not added again. |
| note add | Adds a Markdown note from a file or a string, optionally pinned, and indexes it. |
| glossary import | Upserts accepted terms from a JSON array, each with its own searchable passage. |
| search | Runs the project search the overlay uses, with an optional debug view of why each passage ranked where it did. |
| ask | Asks a question of the project and prints the cited answer, through the AI route you configured in the app. |
| project export | Writes a project out as readable files: originals, transcripts, summaries and a project.json. |
| counts | Prints the row counts of every table, handy for an agent to confirm what it loaded. |
Good to know.
Does my agent see my documents?
Yes: the agent you run reads the files you point it at, under its own privacy terms, exactly as it would read a codebase. That is between you and the agent's provider. meeteor-cli itself sends nothing anywhere: parsing, OCR and indexing happen on your PC.
If you would rather not hand the files to an agent, run the same commands yourself. The agent only saves you the writing.
Which agents work?
Any tool that can run a command on your PC or Mac and read files: Claude Code, Codex, Cursor and the like, or a plain shell script. The CLI has no dependency on any of them. We develop Meeteor with Claude Code, so that is the one we have used most.
Do I need to be a developer?
No, but it is a terminal tool, and the easiest way to use it is to let a coding agent drive it. If you can describe the project in a paragraph, the agent can do the rest.
Can it import transcripts from another tool?
As documents or notes, yes: a transcript file ingests like any text and is searchable. It does not become a Meeteor call, so it has no speaker timeline or action items. Calls come from Meeteor listening.
What happens if I run it twice?
Documents already in the project are skipped (same content, same file), glossary terms merge by name, and the brief is replaced. Notes are added each time, so refresh a status note from the app rather than re-adding it. Take a backup first in any case; it is one command.
Does it need an AI key?
Not for building a project. ingest, note add, glossary import and search run entirely on your PC. ask uses the AI route you configured in the app: Ollama in Local mode, or your own Anthropic or OpenAI key.
Your context, loaded before the call.
Founder licenses are a one-time purchase per major version, up to 75% off, and limited to 100 seats per tier on each platform.