1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
//! TUI event types — messages from the worker thread to the render loop,
//! and user inputs from the render loop back to the worker thread.
/// Events fired by the worker thread and polled by the render loop.
#[derive(Debug)]
pub enum TuiEvent {
/// Streaming text delta — one or more tokens from the model.
Token(String),
/// A full turn completed.
TurnComplete {
text: String,
iterations: u32,
in_tok: u32,
out_tok: u32,
},
/// A tool call started (before the tool runs).
ToolCallStart { name: String, input_preview: String },
/// A tool call finished.
ToolCallDone {
name: String,
result_preview: String,
ok: bool,
elapsed_ms: u64,
},
/// Session was auto-compacted; `removed` messages were summarised.
Compacted { removed: usize },
/// Session was persisted to disk.
Saved,
/// A turn failed with this error message.
TurnError(String),
/// Worker is actively running a turn (true) or idle (false).
Working(bool),
/// Current estimated session token count + compaction threshold.
TokensUpdate { estimated: usize, threshold: usize },
/// Worker rebuilt the runtime from a fresh session (response to /clear).
SessionReset,
/// Informational text from a slash command (e.g. /help, /status, /tools).
/// Rendered as a non-error system message in the chat history.
Info(String),
/// The worker thread is blocked inside a permission prompt for a
/// `DangerFullAccess` tool. The render loop must show the confirmation
/// modal and answer over `resp_tx` — NOT over `UserInput`, because the
/// worker owns that receiver and cannot read it while parked inside
/// `run_turn`.
///
/// Each request carries its own rendezvous channel, so a buffered or
/// stale answer from an earlier prompt can never satisfy a later one,
/// and any render-loop exit path that drops the prompt automatically
/// denies it (the worker's `recv()` sees `Disconnected`).
PermissionRequest {
tool_name: String,
/// The tool's full input — never truncated here. The render side
/// owns presentation (wrap + scroll), so no payload can hide past
/// a preview edge (same rationale as `CliPrompter`).
input: String,
/// Display string of the tier the tool requires
/// (e.g. "danger-full-access").
required_mode: String,
/// Answer channel: `true` → allow, `false` → deny.
resp_tx: std::sync::mpsc::SyncSender<bool>,
},
}
/// One image attached to a user turn — base64-encoded payload paired with
/// its MIME type. Both transports (Ollama `images: [b64,…]` and
/// OpenAI-compat `image_url` data URLs) consume this directly.
#[derive(Debug, Clone)]
pub struct ImageAttachment {
pub media_type: String,
pub data_b64: String,
}
/// Commands sent from the TUI render loop to the worker thread.
#[derive(Debug)]
pub enum UserInput {
/// User submitted a text message to send to the model. `images` is
/// empty for plain-text turns and non-empty when the user pasted /
/// drag-dropped image attachments before pressing Enter.
Message {
text: String,
images: Vec<ImageAttachment>,
},
/// User typed a slash command (e.g. `clear` for `/clear`).
SlashCommand(String),
/// User quit the TUI.
Quit,
}