Terminal UI

quiver-tui browses and manages a running server from the terminal: a dashboard of collections, the points in each, insert and search forms, and snapshots. It talks to the REST API through the Rust client.

cargo run --release -p quiver-tui                 # against http://localhost:6900
./target/release/quiver-tui --url http://db.internal:6900 --api-key "$KEY"

The Docker image includes it too: docker compose exec quiver quiver-tui.

Options

Flag	Environment	Default	Meaning
`--url`	`QUIVER_URL`	`http://localhost:6900`	the server's REST address
`--api-key`	`QUIVER_API_KEY`	none	sent as `Authorization: Bearer <key>`
`--ca-cert`	`QUIVER_CA_CERT`	none	a PEM CA certificate to trust, for an `https://` server with a private CA
`--client-cert`	`QUIVER_CLIENT_CERT`	none	a PEM client certificate followed by its key, for mutual TLS
`--embedding-url`	`QUIVER_EMBEDDING_URL`	none	an OpenAI-compatible embeddings endpoint; enables text input
`--embedding-model`	`QUIVER_EMBEDDING_MODEL`	none	the model name to send
`--embedding-api-key`	`QUIVER_EMBEDDING_API_KEY`	none	a bearer token for the embeddings endpoint

Unlike the server, quiver-tui does not read a .env file. Export the variables, or pass the flags.

Screens

Dashboard — the server's address, version, storage engine, whether snapshots are on, and its totals; below, every collection with its dimension, metric, index, points, tombstones and other (named and sparse) vectors.
Collection — the collection's points, 50 to a page, in id order, with a preview of their metadata. After a search, the results with their distances; Esc goes back to the points.
Snapshots — every snapshot, with its collection, point count, creation time and size.

Enter on a point or a result opens its vector and metadata. Destructive actions (dropping a collection, deleting a point or a snapshot) ask for y first.

Key bindings

? shows them all.

Where	Key	Does
Anywhere	`?`	help
	`r`	refresh
	`q` / `Ctrl`+`C`	quit
	`Esc`	back, or close a dialog (quits from the dashboard)
Dashboard	`↑` `↓` / `k` `j`	select a collection
	`Enter`	open it
	`n`	new collection
	`d`	drop the selected collection
	`s`	snapshots
Collection	`PgDn` / `→`, `PgUp` / `←`	next / previous page
	`Enter`	point details
	`i`	insert a point
	`/`	search
	`x`	delete the selected point
	`S`	snapshot this collection
Snapshots	`c`	snapshot the collection selected on the dashboard
	`R`	restore the selected snapshot
	`D`	delete it
Forms	`Tab` / `↓`, `Shift`+`Tab` / `↑`	next / previous field
	`←` `→` / `Space`	change a choice; `Space` flips a toggle
	`Ctrl`+`T`	switch between vector and text input
	`Enter` / `Esc`	submit / cancel

Forms

New collection — name, dimension, metric and index (hnsw, flat or kd_tree, with default parameters).
Insert — an id (blank for a generated UUID), the vector as comma-separated numbers or text to embed, and metadata as a JSON object. It sets only the main vector, so collections with named vectors need another client to insert.
Search — the query, k (default 10) and an optional JSON filter. It searches the main vector.
Restore — a new name (blank for the original) and whether to replace an existing collection.

Invalid input, such as a non-number in a vector or metadata that is not a JSON object, is reported in the form before anything is sent; server errors are shown in the status line.

Text embeddings

With --embedding-url, the insert and search forms also accept text: Ctrl+T switches the input to text, and the TUI turns it into a vector with the endpoint before inserting or searching. It sends {"input": "<text>", "model": "<model>"} and reads data[0].embedding from the answer, the OpenAI embeddings format, so OpenAI, Ollama and other compatible servers work. With Ollama:

ollama pull nomic-embed-text
quiver-tui --embedding-url http://localhost:11434/v1/embeddings --embedding-model nomic-embed-text

The collection's dimension must match the model's: nomic-embed-text returns 768 numbers, so create the collection with dimension 768. Without --embedding-url, Ctrl+T only explains how to enable it.