| Commit message (Collapse) | Author | Age |
|---|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
This PR allows overriding the Atuin AI model for feature-flagged users
with the `ATUIN_AI__MODEL` environment variable. Any OpenRouter model is
accepted; it's recommended to include an explicit `openrouter:` provider
prefix, so models with colons in their names are interpreted correctly,
e.g. `ATUIN_AI__MODEL=openrouter:deepseek/deepseek-v4-flash:free`.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Fixes keystroke lag in Atuin AI that scales with conversation length.
After extended use (many turns, lots of tool calls with output
viewports), pressing a key had noticeable delay before the letter
appeared.
Three layers of optimization:
- **Skip `sync_view_state` for `InputUpdated`** — every keystroke was
cloning all events, tools, and archived data even though no FSM state
changed. Uses `handle.update_tracked()` (eye_declare 0.5) to skip
rebuilds when values haven't actually changed.
- **Pre-compute turns and `has_command` on the driver thread** — the
view function was rebuilding the full turn structure from raw events and
scanning for `suggest_command` tool calls 3× per render frame. Now
computed once per FSM state change and cached in ViewState.
- **Commit-based element tree pruning** — turns that scroll into
terminal scrollback are tracked via `on_commit` and filtered from the
element tree, keeping rendering work proportional to visible content.
Turn views are now direct children of the root VStack (not nested inside
AtuinAi) so `detect_committed` can see them.
|
| |
|
| |
Adds a skills system that lets users define reusable LLM instructions as `SKILL.md` files with YAML frontmatter.
|
| |
|
|
|
|
|
|
|
|
| |
This PR adds the ability to inject user-defined content into Atuin AI
requests, a la `AGENTS.md` or `CLAUDE.md`.
* `.atuin/TERMINAL.md` (or alternatively just `TERMINAL.md`) is checked in every directory from the cwd up to the root
* `~/.config/atuin/TERMINAL.md` (or equivalent config dir) is also
checked
* Supports Claude-style ``` !`` ``` and ```` ```!...``` ```` style shell
interpolation
|
| |
|
|
|
|
|
|
|
|
| |
This PR fixes two minor issues when using Atuin with fish's vim mode(s):
1. Add support for `fish_hybrid_key_bindings`: in addition to
`fish_vi_key_bindings`, fish also provides `fish_hybrid_key_bindings`,
which is a vi-based mode with some extra keybinds; but the key bind
check for `atuin search` only checks for the former.
2. The AI keybind inserts a `?` on the command line in vim mode even
though it can only be called from normal mode, when pressing `?`
normally does nothing.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
Replaces the tangled dispatch handler system (`tui/dispatch.rs`,
`tui/state.rs`) with a pure finite state machine + driver architecture.
The FSM handles all state transitions as explicit `(State, Event) →
(NewState, Effects)` mappings. The driver executes IO effects and
bridges the TUI to the FSM.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
## Summary
Implements the `write_file` client-side tool — creates new files or
overwrites existing ones with an explicit `overwrite` flag for safety.
- **Overwrite flag**: Writing to an existing file without `overwrite:
true` returns an error directing the LLM to set the flag or use
`edit_file` for targeted changes. Prevents accidental overwrites.
- **Snapshots**: Existing files are backed up before overwriting (same
infrastructure as `edit_file`).
- **Content preview**: Completed writes show the first 10 lines in gray
with line numbers, plus "+ N more lines" for longer files.
- **Atomic writes**: Uses `tempfile` + fsync + rename (same as
`edit_file`).
- **File tracker update**: After writing, the file is registered in the
tracker so subsequent `edit_file` calls work without a separate read.
- **Permission**: Shares the `"Write"` rule with `edit_file` — one
permission covers both tools.
|
| |
|
|
|
|
| |
Overhaul of how AI tool calls are modeled, rendered, and displayed in
the Atuin AI TUI. Fixes bugs in shell command output capture, implements
the `edit_file` tool with full safety infrastructure, and adds a diff
preview for edits.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
<!-- Thank you for making a PR! Bug fixes are always welcome, but if
you're adding a new feature or changing an existing one, we'd really
appreciate if you open an issue, post on the forum, or drop in on
Discord -->
## Checks
- [ ] I am happy for maintainers to push small adjustments to this PR,
to speed up the review cycle
- [ ] I have checked that there are no existing pull requests for the
same thing
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This PR introduces session continuation to Atuin AI.
* Conversations with Atuin AI are stored in a local SQLite database
* Upon startup, Atuin AI tries to find a session to resume based on its
directory/workspace and the time since the last event
* If found, Atuin AI will show a note that the session has been resumed,
and an event is added to help the LLM know where the invocation
boundaries are
* If not, Atuin AI will create a new conversation
* The user can create a new conversation with `/new`
* The new setting `ai.session_continue_minutes`, which defaults to `60`,
controls how old the last event in a session can be before it's no
longer considered for automatic resuming.
<img width="1055" height="593" alt="image"
src="https://github.com/user-attachments/assets/3f9ff01a-ef64-44a9-b0e2-3a4252c5746f"
/>
## Architecture
A new `SessionService` trait defines an API contract for a service that
can manage session data. `LocalSessionService` implements this, with
`DaemonSessionService` a possible future extension point.
`SessionManager` owns a `dyn SessionService` and delegates as
appropriate.
|
| |
|
|
| |
(#3404)
|
| |
|
|
|
|
| |
Adds client-side tool execution to Atuin AI, starting with
`atuin_history`. The server can request tool calls, which are executed
locally with a permission system, and results are sent back to continue
the conversation.
|
| |
|
|
|
|
|
|
|
|
| |
This enables it to perform more effectively and give better suggestions.
Same as send_cwd, disabled by default, opt in.
## Checks
- [ ] I am happy for maintainers to push small adjustments to this PR,
to speed up the review cycle
- [ ] I have checked that there are no existing pull requests for the
same thing
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
eye-declare v0.2.0 includes two relevant changes:
* "Capture-phase" event handling lets us remove the special-case event
handling in `InputBox` that allowed global keyboard binds to work; the
`AtuinAi` component now handles this directly.
* `Tracked::read()` allows reading fields in tracked state without
triggering the dirty-tracking from the `DerefMut` implementation,
allowing us to send messages on the app event bus without marking the
containing state as dirty.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This PR replaces the mess of custom rendering code in Atuin AI with
[eye-declare](https://github.com/BinaryMuse/eye-declare), and updates
the TUI to feel more terminal-native: output appears inline and persists
in scrollback, so you can scroll up and look at previous conversations
for reference.
The "review" state — which used to exist between the LLM generating a
response and the user either executing or following up — has been
removed; just start typing to follow up with the LLM, or press `enter`
at the empty input box to execute the suggested command.
<img width="1203" height="633" alt="image"
src="https://github.com/user-attachments/assets/159ee447-9a2a-4edd-b56e-a79bf1aaaa94"
/>
|
| |
|
|
|
|
|
| |
This PR updates `atuin init` to skip initializers for subfeatures when
that subfeature is explicitly disabled with a setting value of `false`.
For `ai.enabled = false`, this releases the question mark keybind.
Fixes #3325
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is the equivalent of #3208, but for Windows.
~The rendering performance is noticeably slower in this mode when
refreshing a large part of the screen, but it's better than not having
the feature at all.~ Fixed 🙂
The second commit fixes some unrelated warnings.
/cc @BinaryMuse FYI
## Checks
- [x] I am happy for maintainers to push small adjustments to this PR,
to speed up the review cycle
- [x] I have checked that there are no existing pull requests for the
same thing
|
| |
|
|
|
|
|
|
|
|
|
|
| |
<!-- Thank you for making a PR! Bug fixes are always welcome, but if
you're adding a new feature or changing an existing one, we'd really
appreciate if you open an issue, post on the forum, or drop in on
Discord -->
## Checks
- [ ] I am happy for maintainers to push small adjustments to this PR,
to speed up the review cycle
- [ ] I have checked that there are no existing pull requests for the
same thing
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I discovered that since #3178, typing or pasting a `?` in a URL no
longer gets escaped by `url-quote-magic`. For example, pasting
`https://example.com/search?q=foo&test` would result in
`https://example.com/search?q=foo\&test` (leave the `?` unescaped, while
`&` still worked correctly).
The root cause is that Atuin binds `?` to `_atuin_ai_question_mark`,
which bypasses `url-quote-magic` in two ways:
1. **Typed `?`**: the else branch (non-empty buffer) appended `?`
directly to `LBUFFER` instead of delegating to `self-insert` which runs
`url-quote-magic`.
2. **Pasted `?`**: `bracketed-paste-magic` only [activates widgets whose
name matches
`self-*`](https://github.com/zsh-users/zsh/blob/99f578897614f318cdad76402a7d2423ce176b5a/Functions/Zle/bracketed-paste-magic#L24).
Since `_atuin_ai_question_mark` didn't match, pasted `?` characters fell
through to `zle .self-insert` — the raw built-in that inserts literally
without any URL escaping.
The fix renames the widget to `self-atuin-ai-question-mark` (Note: I am
not sure this is the best way but it is a relatively simple one). The
`self-` prefix satisfies `bracketed-paste-magic`'s `active-widgets`
pattern, so `?` in paste is processed by our widget and delegated to
`zle self-insert`, restoring `url-quote-magic` behaviour. The typed case
delegates to `zle self-insert` in the else branch for the same reason.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
| |
* Run Atuin AI's `init` during main `init` for bash, zsh, and fish
* Note that logging into Hub will enable sync
* Add instructions for users with existing sync accounts
* Ensure daemon respects `auto_sync` setting
* Update docs on disabling Atuin AI
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
<!-- Thank you for making a PR! Bug fixes are always welcome, but if
you're adding a new feature or changing an existing one, we'd really
appreciate if you open an issue, post on the forum, or drop in on
Discord -->
Listen to signals as well as state
## Checks
- [ ] I am happy for maintainers to push small adjustments to this PR,
to speed up the review cycle
- [ ] I have checked that there are no existing pull requests for the
same thing
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
## Summary
This PR enables the Atuin CLI to authenticate with Atuin Hub, unifying
authentication across CLI sync and Hub features (AI, runbooks, etc.).
### Key Changes
- **Dual auth support**: New `AuthToken` enum supports both `Bearer`
(Hub) and `Token` (legacy CLI) authentication
- **Smart protocol selection**: New `sync_protocol` setting
(`auto`/`hub`/`legacy`) determines auth method. By default,
`api.atuin.sh` uses Hub auth; custom sync addresses use legacy auth
- **Hub login flow**: `atuin login` now initiates an OAuth-like flow for
Hub users—generates a code, user authorizes in browser, CLI polls for
completion
- **Account linking**: After Hub auth, silently attempts to link
existing CLI sync account to Hub account for seamless migration
- **Graceful fallback**: `sync_auth_token()` prefers Hub token when
available, falls back to CLI session token
### Auth Flow
1. User runs `atuin login` (with default sync address)
2. CLI requests auth code from Hub, displays URL
3. User opens URL, logs in/registers on Hub
4. Hub attaches API token to code
5. CLI polls, receives token, saves as hub session
6. If user had existing CLI sync account, it's automatically linked
### Backward Compatibility
- Existing self-hosted users: unaffected (legacy auth via `Token`
header)
- Existing `api.atuin.sh` users: continue working with CLI session until
they run `atuin login`
- New users: go through Hub flow automatically
## Test Plan
- [ ] New user registration via Hub flow
- [ ] Existing CLI user can still sync without changes
- [ ] `atuin login` links CLI account to Hub account
- [ ] Self-hosted users unaffected by changes
- [ ] AI commands work after Hub auth
---------
Co-authored-by: Ellie Huxtable <ellie@elliehuxtable.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(#3234)
It feels much, much nicer this way. This has also been asked for pretty
consistently since we made inline rendering the default. Now we can have
everything :)
Maintains a shadow vt100 renderer so that we can restore the terminal
state upon popup close. This happens on a background thread, so our
impact on terminal performance should still be super minimal, if
anything
## Checks
- [ ] I am happy for maintainers to push small adjustments to this PR,
to speed up the review cycle
- [ ] I have checked that there are no existing pull requests for the
same thing
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
## Summary
This PR adds a persistent, in-memory search index to the Atuin daemon,
enabling fast fuzzy search without the startup delay of building an
index each time the TUI opens.
### Key Changes
- **Daemon search service**: A new gRPC service that maintains a Nucleo
fuzzy search index in memory
- **Real-time index updates**: The daemon listens for history events
(new commands, synced records) and updates the index immediately
- **Filter mode support**: All existing filter modes work (Global, Host,
Session, Directory, Workspace)
- **New search engine**: `daemon-fuzzy` search mode that queries the
daemon instead of building a local index
- **Paged history loading**: Database pagination support for efficient
initial index loading
- **Configurable logging**: New `[logs]` settings section for daemon and
search log configuration
- **Component-based daemon architecture**: Refactored daemon internals
into a modular, event-driven system
- **Fallback to DB search for regex**: Since Nucleo doesn't support
regex matching
## Daemon Architecture
The daemon has been refactored to use a component-based, event-driven
architecture that makes it easier to add new functionality and reason
about the system.
### Core Concepts
```
┌─────────────────────────────────────────────────────────────────────────────┐
│ Atuin Daemon │
│ │
│ ┌─────────────┐ ┌──────────────────────────────────────────────────┐ │
│ │ Daemon │ │ Components │ │
│ │ Handle │────▶│ │ │
│ │ │ │ ┌─────────────┐ ┌─────────────┐ ┌────────────┐ │ │
│ │ • emit() │ │ │ History │ │ Search │ │ Sync │ │ │
│ │ • subscribe │ │ │ Component │ │ Component │ │ Component │ │ │
│ │ • settings │ │ │ │ │ │ │ │ │ │
│ │ • databases │ │ │ gRPC service│ │ gRPC service│ │ background │ │ │
│ └─────────────┘ │ │ WIP history │ │ Nucleo index│ │ sync │ │ │
│ │ │ └─────────────┘ └─────────────┘ └────────────┘ │ │
│ │ └──────────────────────────────────────────────────┘ │
│ │ ▲ │
│ ▼ │ │
│ ┌─────────────────────────────────────┴────────────────────────────────┐ │
│ │ Event Bus (broadcast) │ │
│ │ │ │
│ │ HistoryStarted │ HistoryEnded │ RecordsAdded │ SyncCompleted │ ... │ │
│ └──────────────────────────────────────────────────────────────────────┘ │
│ ▲ │
│ ┌───────────────────────────────────┴──────────────────────────────────┐ │
│ │ Control Service (gRPC) │ │
│ │ External event injection from CLI commands │ │
│ └──────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
```
### DaemonHandle
A lightweight, cloneable handle that provides access to shared daemon
resources:
- **Event emission**: `handle.emit(DaemonEvent::...)` broadcasts to all
components
- **Event subscription**: `handle.subscribe()` returns a receiver for
the event bus
- **Settings**: `handle.settings()` for configuration access
- **Databases**: `handle.history_db()` and `handle.store()` for data
access
### Component Trait
Components implement a simple lifecycle:
```rust
#[async_trait]
trait Component: Send + Sync {
fn name(&self) -> &'static str;
async fn start(&mut self, handle: DaemonHandle) -> Result<()>;
async fn handle_event(&mut self, event: &DaemonEvent) -> Result<()>;
async fn stop(&mut self) -> Result<()>;
}
```
### Event-Driven Design
Components communicate via events rather than direct coupling:
| Event | Emitted By | Consumed By |
|-------|-----------|-------------|
| `HistoryStarted` | History gRPC | Search (logging) |
| `HistoryEnded` | History gRPC | Search (index update) |
| `RecordsAdded` | Sync | Search (index update) |
| `HistoryPruned` | CLI (via Control) | Search (index rebuild) |
| `HistoryDeleted` | CLI (via Control) | Search (index rebuild) |
| `ForceSync` | CLI (via Control) | Sync |
| `ShutdownRequested` | Signal handler | All (graceful shutdown) |
### External Event Injection
CLI commands can inject events into a running daemon:
```rust
// After `atuin history prune`
emit_event(DaemonEvent::HistoryPruned).await?;
// After deleting specific items
emit_event(DaemonEvent::HistoryDeleted { ids }).await?;
// Request immediate sync
emit_event(DaemonEvent::ForceSync).await?;
```
This ensures the daemon's search index stays in sync with database
changes made by CLI commands.
## Search Architecture
The search service uses a [forked version of
Nucleo](https://github.com/atuinsh/nucleo-ext) that adds filter and
scorer callbacks, enabling efficient filtering and frecency-based
ranking.
```
┌────────────────────────────────────────────────────────────────┐
│ Atuin Daemon │
│ │
│ ┌─────────────────┐ ┌──────────────────────────────────┐ │
│ │ Event System │───▶│ Search Component │ │
│ │ │ │ │ │
│ │ • RecordsAdded │ │ ┌────────────────────────────┐ │ │
│ │ • HistoryEnded │ │ │ Deduplicated Index │ │ │
│ │ • HistoryPruned │ │ │ │ │ │
│ └─────────────────┘ │ │ CommandData per command: │ │ │
│ │ │ • Global frecency │ │ │
│ ┌─────────────────┐ │ │ • Filter indexes (sets) │ │ │
│ │ Background Task │ │ │ • Invocation history │ │ │
│ │ │ │ └────────────────────────────┘ │ │
│ │ Rebuilds │ │ │ │ │
│ │ frecency map │ │ ▼ │ │
│ │ every 60s │───▶│ ┌────────────────────────────┐ │ │
│ └─────────────────┘ │ │ Nucleo (forked) │ │ │
│ │ │ │ │ │
│ │ │ • Filter callback │ │ │
│ │ │ • Scorer callback │ │ │
│ │ │ • Fuzzy matching │ │ │
│ │ └────────────────────────────┘ │ │
│ └──────────────────────────────────┘ │
│ │ │
│ │ gRPC (Unix socket) │
└──────────────────────────────────────│─────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────┐
│ Search TUI (Client) │
│ │
│ 1. Send query + filter mode + context to daemon │
│ 2. Receive matching history IDs (ranked by frecency) │
│ 3. Hydrate full records from local SQLite database │
│ 4. Display results in TUI │
└─────────────────────────────────────────────────────────────────┘
```
### Nucleo Fork
The [nucleo-ext fork](https://github.com/atuinsh/nucleo-ext) adds two
key features to Nucleo:
1. **Filter callback**: Pre-filter items before fuzzy matching (used for
directory/host/session filtering)
2. **Scorer callback**: Compute custom scores after matching (used for
frecency ranking)
```rust
// Filter: only include commands run in current directory
nucleo.set_filter(Some(Arc::new(|cmd: &String| {
passing_commands.contains(cmd)
})));
// Scorer: combine fuzzy score with frecency
nucleo.set_scorer(Some(Arc::new(|cmd: &String, fuzzy_score: u32| {
let frecency = frecency_map.get(cmd).unwrap_or(0);
fuzzy_score + (frecency * 10)
})));
```
### Deduplicated Index
Commands are stored once per unique command text, with metadata tracking
all invocations:
```rust
struct CommandData {
command: String,
invocations: Vec<Invocation>, // All times this command was run
global_frecency: FrecencyData, // Precomputed frecency score
// O(1) filter indexes
directories: HashSet<String>, // All cwds where command was run
hosts: HashSet<String>, // All hostnames
sessions: HashSet<String>, // All session IDs
}
```
This deduplication means:
- **Fewer items to match**: ~13K unique commands vs ~62K history entries
- **O(1) filter checks**: HashSet lookups instead of scanning
invocations
- **Single frecency score**: Global frecency computed once, used for all
filter modes
### Frecency Scoring
Frecency (frequency + recency) scoring prioritizes recently and
frequently used commands:
```rust
fn compute_frecency(count: u32, last_used: i64, now: i64) -> u32 {
let age_hours = (now - last_used) / 3600;
// Recency: decays over time (half-life ~24 hours)
let recency = (100.0 * (-age_hours as f64 / 24.0).exp()) as u32;
// Frequency: logarithmic scaling
let frequency = (count.ln() * 20.0).min(100.0) as u32;
recency + frequency
}
```
The frecency map is:
- **Precomputed by background task** every 60 seconds
- **Never computed inline** during search (no latency impact)
- **Graceful fallback**: If unavailable, search works without frecency
ranking
### Filter Mode Implementation
| Filter Mode | Implementation |
|-------------|----------------|
| Global | No filter (all commands) |
| Directory | `command.directories.contains(cwd)` |
| Workspace | `command.directories.any(\|d\| d.starts_with(git_root))` |
| Host | `command.hosts.contains(hostname)` |
| Session | `command.sessions.contains(session_id)` |
Filters are pre-computed into a HashSet before the search, making the
filter callback O(1).
### Search Flow
1. **Daemon startup**: Loads history from SQLite in pages, builds
deduplicated index
2. **Frecency precompute**: Background task builds frecency map after
history loads
3. **Search request**: Client sends query with filter mode and context
4. **Filter**: Pre-computed HashSet determines which commands pass the
filter
5. **Match**: Nucleo fuzzy matches the query against command text
6. **Score**: Frecency scorer ranks results (fuzzy score + frecency *
10)
7. **Response**: Returns history IDs for the most recent invocation of
each matching command
8. **Hydration**: Client fetches full records from local SQLite
### Configuration
```toml
# Enable daemon + autostart
[daemon]
enabled = true
autostart = true
# Enable daemon-based fuzzy search
[search]
search_mode = "daemon-fuzzy"
```
## Performance
Performance varies based on several factors, but in most initial testing
with the new architecture shows improvement:
* **Nucleo performs searches up to 4.5x faster**: direct DB search
averages 18.07ms, but the daemon completes the same queries in 3.99ms.
* **IPC overhead is significant, but acceptable**: a significant amount
of wall-time is taken up by the transfer of data over IPC (via UDS in
this case). This averages to about ~7.8ms and accounts for 66% of
client-side wall time.
* **Tail latency improves at every layer**: p99 times correspond to
initial requests, worst-case query patterns, etc. but the average p99
daemon-based response time is 3.6x better than the associated DB-based
search p99 time
* **Query complexity no longer impacts performance**: the Nucleo-based
search shows consistent 2-7ms times regardless of query pattern. The
DB-based search had a 17x variance (3.59ms to 62.46ms).
Interestingly, @ellie - who has a larger history store than I do - gets
even better performance on the IPC layer. This could use a lot more
testing in various edge cases and on various hardware, but seems
promising.
### Regular DB search
```
Individual calls for: db_search
--------------------------------------------------------------------------------------------------------------
# Wall Busy Idle Fields
--------------------------------------------------------------------------------------------------------------
1 32.25ms 32.20ms 47.70µs {"mode":"Fuzzy","query":"^"}
2 19.48ms 19.40ms 84.20µs {"mode":"Fuzzy","query":"^c"}
3 20.40ms 20.10ms 297.00µs {"mode":"Fuzzy","query":"^ca"}
4 13.07ms 13.00ms 69.90µs {"mode":"Fuzzy","query":"^car"}
5 12.17ms 12.10ms 67.10µs {"mode":"Fuzzy","query":"^carg"}
6 20.78ms 20.70ms 76.60µs {"mode":"Fuzzy","query":"^cargo"}
7 9.15ms 9.10ms 53.20µs {"mode":"Fuzzy","query":"^cargo "}
8 10.24ms 10.00ms 237.00µs {"mode":"Fuzzy","query":"^cargo b"}
9 10.01ms 9.68ms 325.00µs {"mode":"Fuzzy","query":"^cargo bu"}
10 5.89ms 5.83ms 57.20µs {"mode":"Fuzzy","query":"^cargo bui"}
11 8.85ms 8.28ms 568.00µs {"mode":"Fuzzy","query":"^cargo buil"}
12 7.70ms 7.49ms 212.00µs {"mode":"Fuzzy","query":"^cargo build"}
13 3.59ms 3.53ms 57.00µs {"mode":"Fuzzy","query":"^cargo build$"}
14 6.50ms 6.44ms 63.60µs {"mode":"Fuzzy","query":"^cargo "}
15 6.48ms 6.38ms 100.00µs {"mode":"Fuzzy","query":"!"}
16 31.68ms 31.60ms 75.90µs {"mode":"Fuzzy","query":"!g"}
17 62.46ms 62.40ms 58.90µs {"mode":"Fuzzy","query":"!gi"}
18 30.35ms 30.30ms 46.90µs {"mode":"Fuzzy","query":"!git"}
19 53.84ms 53.80ms 40.80µs {"mode":"Fuzzy","query":"!git "}
20 19.24ms 19.20ms 39.70µs {"mode":"Fuzzy","query":"!git c"}
21 22.03ms 22.00ms 34.70µs {"mode":"Fuzzy","query":"!git co"}
22 17.13ms 17.00ms 133.00µs {"mode":"Fuzzy","query":"!git com"}
23 16.14ms 15.90ms 242.00µs {"mode":"Fuzzy","query":"!git comm"}
24 5.11ms 5.08ms 28.60µs {"mode":"Fuzzy","query":"!git commi"}
25 7.31ms 7.26ms 52.70µs {"mode":"Fuzzy","query":"!git commit"}
Summary: 25 calls
Wall: avg=18.07ms, min=3.59ms, max=62.46ms, p50=13.07ms, p99=62.46ms
Busy: avg=17.95ms, min=3.53ms, max=62.40ms, p50=13.00ms, p99=62.40ms
```
### Daemon-based search
**Client**
```
Individual calls for: daemon_search
--------------------------------------------------------------------------------------------------------------
# Wall Busy Idle Fields
--------------------------------------------------------------------------------------------------------------
1 13.05ms 2.55ms 10.50ms {"query":"^"}
2 10.65ms 1.40ms 9.25ms {"query":"^c"}
3 10.72ms 1.18ms 9.54ms {"query":"^ca"}
4 5.54ms 485.00µs 5.06ms {"query":"^car"}
5 15.02ms 1.02ms 14.00ms {"query":"^carg"}
6 9.49ms 840.00µs 8.65ms {"query":"^cargo"}
7 5.53ms 555.00µs 4.97ms {"query":"^cargo "}
8 8.56ms 717.00µs 7.84ms {"query":"^cargo b"}
9 12.34ms 1.24ms 11.10ms {"query":"^cargo bu"}
10 8.38ms 650.00µs 7.73ms {"query":"^cargo bui"}
11 13.07ms 770.00µs 12.30ms {"query":"^cargo buil"}
12 17.11ms 709.00µs 16.40ms {"query":"^cargo build"}
13 15.41ms 907.00µs 14.50ms {"query":"^cargo build$"}
14 8.19ms 665.00µs 7.52ms {"query":"^cargo "}
15 7.98ms 1.72ms 6.26ms {"query":"!"}
16 13.56ms 856.00µs 12.70ms {"query":"!g"}
17 8.11ms 624.00µs 7.49ms {"query":"!gi"}
18 14.57ms 775.00µs 13.80ms {"query":"!git"}
19 14.18ms 779.00µs 13.40ms {"query":"!git "}
20 9.62ms 802.00µs 8.82ms {"query":"!git c"}
21 15.50ms 1.50ms 14.00ms {"query":"!git co"}
22 11.58ms 1.48ms 10.10ms {"query":"!git com"}
23 13.82ms 2.12ms 11.70ms {"query":"!git comm"}
24 17.48ms 2.18ms 15.30ms {"query":"!git commi"}
25 14.81ms 1.71ms 13.10ms {"query":"!git commit"}
Summary: 25 calls
Wall: avg=11.77ms, min=5.53ms, max=17.48ms, p50=12.34ms, p99=17.48ms
Busy: avg=1.13ms, min=485.00µs, max=2.55ms, p50=856.00µs, p99=2.55ms
```
**Daemon**
```
Individual calls for: daemon_search_query
--------------------------------------------------------------------------------------------------------------
# Wall Busy Idle Fields
--------------------------------------------------------------------------------------------------------------
1 1.75ms 250ns 1.75ms {"query":"^","query_id":1}
2 4.58ms 125ns 4.58ms {"query":"^c","query_id":2}
3 4.39ms 250ns 4.39ms {"query":"^ca","query_id":3}
4 2.52ms 125ns 2.52ms {"query":"^car","query_id":4}
5 4.44ms 250ns 4.44ms {"query":"^carg","query_id":5}
6 3.66ms 167ns 3.66ms {"query":"^cargo","query_id":6}
7 2.38ms 84ns 2.38ms {"query":"^cargo ","query_id":7}
8 4.13ms 84ns 4.13ms {"query":"^cargo b","query_id":8}
9 4.40ms 167ns 4.40ms {"query":"^cargo bu","query_id":9}
10 3.87ms 125ns 3.87ms {"query":"^cargo bui","query_id":10}
11 4.36ms 84ns 4.36ms {"query":"^cargo buil","query_id":11}
12 3.96ms 333ns 3.96ms {"query":"^cargo build","query_id":12}
13 4.61ms 167ns 4.61ms {"query":"^cargo build$","query_id":13}
14 4.20ms 209ns 4.20ms {"query":"^cargo ","query_id":14}
15 238.17µs 167ns 238.00µs {"query":"!","query_id":15}
16 4.44ms 125ns 4.44ms {"query":"!g","query_id":16}
17 3.47ms 83ns 3.47ms {"query":"!gi","query_id":17}
18 4.57ms 125ns 4.57ms {"query":"!git","query_id":18}
19 7.15ms 167ns 7.15ms {"query":"!git ","query_id":19}
20 4.27ms 250ns 4.27ms {"query":"!git c","query_id":20}
21 5.19ms 292ns 5.19ms {"query":"!git co","query_id":21}
22 4.29ms 417ns 4.29ms {"query":"!git com","query_id":22}
23 4.08ms 125ns 4.08ms {"query":"!git comm","query_id":23}
24 4.50ms 167ns 4.50ms {"query":"!git commi","query_id":24}
25 4.35ms 208ns 4.35ms {"query":"!git commit","query_id":25}
Summary: 25 calls
Wall: avg=3.99ms, min=238.17µs, max=7.15ms, p50=4.29ms, p99=7.15ms
Busy: avg=182ns, min=83ns, max=417ns, p50=167ns, p99=417ns
```
**Nucleo matching time (in daemon)**
```
Individual calls for: nucleo_match
--------------------------------------------------------------------------------------------------------------
# Wall Busy Idle Fields
--------------------------------------------------------------------------------------------------------------
1 1.73ms 125ns 1.73ms {"query":"^","query_id":1}
2 4.57ms 167ns 4.57ms {"query":"^c","query_id":2}
3 4.37ms 125ns 4.37ms {"query":"^ca","query_id":3}
4 2.51ms 84ns 2.51ms {"query":"^car","query_id":4}
5 4.43ms 125ns 4.43ms {"query":"^carg","query_id":5}
6 3.64ms 125ns 3.64ms {"query":"^cargo","query_id":6}
7 2.37ms 84ns 2.37ms {"query":"^cargo ","query_id":7}
8 4.11ms 125ns 4.11ms {"query":"^cargo b","query_id":8}
9 4.36ms 208ns 4.36ms {"query":"^cargo bu","query_id":9}
10 3.85ms 125ns 3.85ms {"query":"^cargo bui","query_id":10}
11 4.35ms 125ns 4.35ms {"query":"^cargo buil","query_id":11}
12 3.94ms 250ns 3.94ms {"query":"^cargo build","query_id":12}
13 4.59ms 125ns 4.59ms {"query":"^cargo build$","query_id":13}
14 4.18ms 84ns 4.18ms {"query":"^cargo ","query_id":14}
15 220.13µs 125ns 220.00µs {"query":"!","query_id":15}
16 4.43ms 125ns 4.43ms {"query":"!g","query_id":16}
17 3.45ms 125ns 3.45ms {"query":"!gi","query_id":17}
18 4.55ms 125ns 4.55ms {"query":"!git","query_id":18}
19 7.12ms 209ns 7.12ms {"query":"!git ","query_id":19}
20 4.25ms 166ns 4.25ms {"query":"!git c","query_id":20}
21 5.18ms 125ns 5.18ms {"query":"!git co","query_id":21}
22 4.27ms 125ns 4.27ms {"query":"!git com","query_id":22}
23 4.06ms 292ns 4.06ms {"query":"!git comm","query_id":23}
24 4.46ms 166ns 4.46ms {"query":"!git commi","query_id":24}
25 4.31ms 208ns 4.31ms {"query":"!git commit","query_id":25}
Summary: 25 calls
Wall: avg=3.97ms, min=220.13µs, max=7.12ms, p50=4.27ms, p99=7.12ms
Busy: avg=147ns, min=84ns, max=292ns, p50=125ns, p99=292ns
```
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This PR refines the system created in #3178 to be suitable for a v1
release.
---
## Overview
`atuin-ai` is a separate binary that allows for generating commands and
asking questions from the command line.
It is fully opt-in.
## Usage
`atuin ai init` will output bindings for your shell. Currently, bash,
zsh, and fish are supported.
```bash
eval "$(atuin ai init)"
```
Once the hooks are installed, just press `?` on an empty prompt line to
call up the TUI.
`atuin ai` requires an account on [Atuin Hub](https://hub.atuin.sh/);
you will be prompted to log in on first use.
## Features
### Command generation
Prompt the LLM to create a command, and get one back, no fuss. Press
`enter` to run, or `tab` to insert.
```
┌Ask questions or generate a command:──────────────────────────┐
│ │
│ > Get a list of running docker containers │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ $ docker ps │
│ │
└────[Enter]: Run [Tab]: Insert [f]: Follow-up [Esc]: Cancel┘
```
### Follow-up
You can follow-up with `f` to specify a refinement prompt to update the
command that will be inserted.
```
┌Ask questions or generate a command:──────────────────────────┐
│ │
│ > Get a list of running docker containers │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ $ docker ps │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ > Actually I want to get all docker containers │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ $ docker ps -a │
│ │
└────[Enter]: Run [Tab]: Insert [f]: Follow-up [Esc]: Cancel┘
```
You can also follow-up with questions to get responses in natural
language.
```
┌Ask questions or generate a command:──────────────────────────┐
│ │
│ > Get a list of running docker containers │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ $ docker ps │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ > Actually I want to get all docker containers │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ $ docker ps -a │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ > What other useful flags to `docker ps` should I know? │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ Here are some handy `docker ps` flags: │
│ │
│ - `-q` — Only show container IDs (great for piping to │
│ other commands) │
│ - `-s` — Show container sizes │
│ - `-n 5` — Show the last 5 created containers │
│ - `-l` — Show only the latest created container │
│ - `--no-trunc` — Don't truncate output (shows full IDs and │
│ commands) │
│ - `-f` or `--filter` — Filter by condition, e.g.: │
│ - `-f status=exited` — only exited containers │
│ - `-f name=myapp` — filter by name │
│ - `-f ancestor=nginx` — filter by image │
│ - `--format` — Custom output using Go templates, e.g.: │
│ `--format "table {{.Names}}\t{{.Status}}\t{{.Ports}}"` │
│ │
│ A common combo is `docker ps -aq` to get all container │
│ IDs, useful for bulk operations like `docker rm $(docker │
│ ps -aq)`. │
│ │
└────[Enter]: Run [Tab]: Insert [f]: Follow-up [Esc]: Cancel┘
```
You can use `enter` or `tab` at any time to run or insert the last
suggested command, even if it was suggested in a previous turn.
### Conversational and search usage
If you prompt the LLM with a question that doesn't imply you want to
generate a command, it can respond in natural language, and use web
search if necessary to fetch the data it needs.
```
┌Ask questions or generate a command:──────────────────────────┐
│ │
│ > What is the latest version of atuin? │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ ✓ Used 2 tools │
│ │
│ The latest version of Atuin is **v18.12.0**, available on │
│ the [GitHub releases │
│ page](https://github.com/atuinsh/atuin/releases). │
│ │
└─────────────────────────────────[f]: Follow-up [Esc]: Cancel┘
```
### Dangerous or low-confidence command detection
The LLM scores its confidence in the command, as well as how dangerous
the command is. This information is shown if a threshold is exceeded,
and requires an extra confirmation step before running automatically
with `enter`.
The Atuin Hub server also monitors suggested commands for dangerous
patterns the LLM didn't catch, and appends its own assessment at the end
of the LLM's own assessment.
```
┌Ask questions or generate a command:──────────────────────────┐
│ │
│ > Delete all files from $HOME │
│ │
├──────────────────────────────────────────────────────────────┤
│ │
│ $ rm -rf $HOME/* │
│ │
│ ! ⚠️ This will PERMANENTLY delete ALL files and directories │
│ in your home directory, including documents, downloads, │
│ configurations, SSH keys, and everything else. This is │
│ irreversible and will likely break your system. Also note │
│ this won't delete hidden (dot) files — if you want those │
│ too, that's even more destructive.; [Server] Recursive │
│ delete of critical directory │
│ │
└────[Enter]: Run [Tab]: Insert [f]: Follow-up [Esc]: Cancel┘
```
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
|
| |
|
