Make model and permission choices provider-specific

This commit is contained in:
iris committed 2026-09-08 00:08:09 -04:00
1 parent 6a0202b1b5
commit 7ee88dfd9c
10 files changed
+269 -60

No files matched your search

+7
View File
@@ -221,6 +221,13 @@ phone and auto-resume need no Codex branch. This is the CLI's local protocol
and is treated defensively for the same reason as Claude's undocumented usage
endpoint: missing fields or a refusal degrade to an unavailable snapshot.
The model picker asks the selected setup's Codex app-server for `model/list`
when it is opened. The catalog is account- and CLI-version-specific, so it is
never copied into the app or inferred from another setup; a lookup failure is
shown as unavailable while the free-text escape remains. Permission choices
are likewise reported per provider: Codex offers its read-only,
workspace-write and full-access modes, while Claude keeps its own modes.
### The llama driver
One `llama-server` per session, started through the same `Transport` as any
@@ -330,7 +330,13 @@ fun deleteSubagents(settings: ServerSettings, sessionId: String, subagentIds: Li
//
// One list rather than two. A provider only exists on a machine that has it installed, so offering
// machines and providers as independent choices would offer pairs that cannot work.
data class Provider(val name: String, val kind: String, val models: List<String>)
data class Provider(
val name: String,
val kind: String,
val models: List<String>,
val permissionModes: List<String>,
val defaultPermissionMode: String?,
)
/**
* A machine, and what it can run. [address] is absent for the backend itself.
@@ -345,13 +351,17 @@ data class Setup(
val providers: List<Provider>,
)
private fun parseProvider(provider: JSONObject) =
Provider(
private fun parseProvider(provider: JSONObject): Provider {
val kind = provider.getString("kind")
return Provider(
name = provider.getString("name"),
kind = provider.getString("kind"),
kind = kind,
// Omitted entirely when the provider offers none.
models = provider.optJSONArray("models")?.strings().orEmpty(),
permissionModes = provider.optJSONArray("permissionModes")?.strings().orEmpty(),
defaultPermissionMode = provider.optString("defaultPermissionMode").ifEmpty { null },
)
}
private fun parseSetup(setup: JSONObject) =
Setup(
@@ -1083,16 +1093,6 @@ fun setSessionModel(settings: ServerSettings, sessionId: String, model: String)
) {}
}
/**
* Common permission intents, in the order they give up asking. Each coding CLI maps these onto its
* own flags: "manual" asks, "acceptEdits" and "auto" allow ordinary work, "bypassPermissions"
* accepts everything, and "plan" makes no changes.
*
* One list for every screen that offers them -- spawn, import, and the session's own picker --
* because three copies had already drifted: the import screen was missing "plan".
*/
val PERMISSION_MODES = listOf("manual", "acceptEdits", "auto", "bypassPermissions", "plan")
/**
* What a new session's thinking level is when nothing chose one, or null for the CLI's own.
*
@@ -1294,6 +1294,19 @@ fun fetchSetupModels(settings: ServerSettings, setupId: String): List<LocalModel
}
}
/** The current model catalog for one CLI provider on the machine where it runs. */
fun fetchProviderModels(
settings: ServerSettings,
setupId: String,
provider: String,
): List<String> =
requestFromServer(
settings,
"/setups/${setupId.urlEncoded()}/providers/${provider.urlEncoded()}/models",
) { connection ->
JSONArray(connection.inputStream.bufferedReader().readText()).strings()
}
fun fetchModels(settings: ServerSettings): Models =
requestFromServer(settings, "/models") { connection ->
val body = JSONObject(connection.inputStream.bufferedReader().readText())
@@ -100,9 +100,8 @@ fun ImportScreen(settings: ServerSettings, reloadToken: Int, onImported: (Sessio
// Deleting a transcript cannot be undone, so it is asked rather than done. Held as the rows
// themselves, not a flag, so the dialog can say what it is about.
var confirming by remember { mutableStateOf<List<Importable>?>(null) }
// Same default as the spawn screen: a phone is the wrong place to answer "allow Bash?" forty
// times.
var permissionMode by remember { mutableStateOf("auto") }
// Set from the selected Claude provider rather than repeated in the app.
var permissionMode by remember { mutableStateOf("") }
// When each row last slid upwards, as a plain map rather than state: nothing is drawn from it,
// so a tap reading it needs no recomposition.
val movedAt = remember { mutableMapOf<String, Long>() }
@@ -208,6 +207,9 @@ fun ImportScreen(settings: ServerSettings, reloadToken: Int, onImported: (Sessio
}
val provider = chosen?.providers?.firstOrNull { it.kind == "claude_cli" }
LaunchedEffect(chosen?.id, provider?.name) {
permissionMode = provider?.defaultPermissionMode.orEmpty()
}
/** Continues [targets] in the background, leaving the screen where it is. */
fun importAll(targets: List<Importable>) {
@@ -375,7 +377,7 @@ fun ImportScreen(settings: ServerSettings, reloadToken: Int, onImported: (Sessio
} else {
ChipGroup(
label = "Permissions",
options = PERMISSION_MODES,
options = provider?.permissionModes.orEmpty(),
selected = permissionMode,
onSelect = { permissionMode = it },
)
@@ -305,6 +305,7 @@ fun SessionScreen(
// The models this provider actually offers, asked of the server rather than listed here: a
// hardcoded list is a claim about a machine.
var offeredModels by remember { mutableStateOf<List<String>>(emptyList()) }
var offeredPermissionModes by remember { mutableStateOf<List<String>>(emptyList()) }
val lifecycleOwner = LocalLifecycleOwner.current
// The resume cursor, written from the stream's IO thread.
val lastSeq = remember { AtomicLong(0) }
@@ -1017,22 +1018,28 @@ fun SessionScreen(
// Only for the model picker, which a subagent does not have.
if (!isSubagent) {
LaunchedEffect(summary.setupName, summary.provider) {
offeredModels =
try {
withContext(Dispatchers.IO) {
fetchSetups(settings)
.firstOrNull { it.name == summary.setupName }
?.providers
?.firstOrNull { it.name == summary.provider }
?.models
.orEmpty()
}
} catch (_: Exception) {
// Not worth reporting: the picker simply has nothing to offer, which is
// visible.
emptyList()
LaunchedEffect(summary.setup, summary.provider) {
val provider = runCatching {
withContext(Dispatchers.IO) {
fetchSetups(settings)
.firstOrNull { it.id == summary.setup }
?.providers
?.firstOrNull { it.name == summary.provider }
}
}
.getOrNull()
offeredPermissionModes = provider?.permissionModes.orEmpty()
offeredModels =
provider
?.let {
runCatching {
withContext(Dispatchers.IO) {
fetchProviderModels(settings, summary.setup, summary.provider)
}
}
.getOrDefault(emptyList())
}
.orEmpty()
}
}
@@ -1840,13 +1847,21 @@ fun SessionScreen(
},
)
}
PickerButton(
current = permissionMode,
options = PERMISSION_MODES,
onPick = { chosen ->
act { setSessionPermissionMode(settings, summary.id, chosen) }
},
)
if (offeredPermissionModes.isNotEmpty()) {
PickerButton(
current = permissionMode,
options = offeredPermissionModes,
onPick = { chosen ->
act {
setSessionPermissionMode(
settings,
summary.id,
chosen,
)
}
},
)
}
}
// The same filled shape as the button beside it, not an outlined one: these
// are two things you can do about the session, and weighting one as
@@ -57,10 +57,12 @@ fun SpawnScreen(
var providerName by remember { mutableStateOf<String?>(null) }
var title by remember { mutableStateOf("") }
var model by remember { mutableStateOf("") }
var providerModels by remember { mutableStateOf<List<String>>(emptyList()) }
var providerModelsLoading by remember { mutableStateOf(false) }
var providerModelsError by remember { mutableStateOf<String?>(null) }
var cwd by remember { mutableStateOf("") }
// "auto" rather than "manual": on a phone every ask is a round trip to a question card, and
// answering "allow Bash?" dozens of times per task is what this app exists to avoid.
var permissionMode by remember { mutableStateOf("auto") }
// Set only after the selected provider reports its own default. An empty value is not sent.
var permissionMode by remember { mutableStateOf("") }
// Null until the server has been asked, and null again if it answers "no level chosen" -- the
// two are told apart by [defaultsAsked], because a picker that shows a level before the answer
// arrives is one you can spawn at without having chosen it.
@@ -70,10 +72,8 @@ fun SpawnScreen(
// Only the spawn's own failure. The fetch's lives in `options`: this one leaves a filled-in
// form worth keeping, and that one leaves nothing to fill in.
var spawnError by remember { mutableStateOf<String?>(null) }
// The models on the *chosen machine*, for a llama provider to choose between. Kept separate
// from the setups: a Claude session needs none, so failing to list them must not stop the
// screen rendering. Refetched when the machine changes, because a model is a file on one
// machine -- see [fetchSetupModels].
// GGUFs on the chosen machine, for a llama provider to choose between. Kept separate from a
// coding CLI's provider catalog and refetched when the machine changes.
var models by remember { mutableStateOf<List<LocalModel>>(emptyList()) }
var modelKey by remember { mutableStateOf<String?>(null) }
var contextSize by remember { mutableStateOf("") }
@@ -145,6 +145,28 @@ fun SpawnScreen(
val isCodingCli = isClaude || isCodex
val isLlama = current?.kind == "llama_cpp"
LaunchedEffect(setup?.id, current?.name) {
model = ""
providerModels = emptyList()
providerModelsError = null
permissionMode = current?.defaultPermissionMode.orEmpty()
if (isCodingCli) {
providerModelsLoading = true
try {
providerModels =
withContext(Dispatchers.IO) {
fetchProviderModels(settings, setup.id, current.name)
}
} catch (e: ApiException) {
providerModelsError = e.message
} finally {
providerModelsLoading = false
}
} else {
providerModelsLoading = false
}
}
// The machine first, because it decides what can be run at all.
ChipGroup(
label = "Setup",
@@ -240,14 +262,34 @@ fun SpawnScreen(
}
if (isCodingCli) {
if (current.models.isNotEmpty()) {
Spacer(Modifier.height(16.dp))
ChipGroup(
label = "Model",
options = current.models,
selected = model.ifEmpty { null },
onSelect = { chosen -> model = if (model == chosen) "" else chosen },
)
when {
providerModelsLoading ->
Text(
"Loading model choices…",
style = MaterialTheme.typography.bodySmall,
color = MaterialTheme.colorScheme.onSurfaceVariant,
)
providerModelsError != null ->
Text(
"Model choices unavailable: $providerModelsError",
style = MaterialTheme.typography.bodySmall,
color = MaterialTheme.colorScheme.error,
)
providerModels.isEmpty() ->
Text(
"This setup reported no selectable models.",
style = MaterialTheme.typography.bodySmall,
color = MaterialTheme.colorScheme.onSurfaceVariant,
)
else -> {
Spacer(Modifier.height(16.dp))
ChipGroup(
label = "Model",
options = providerModels,
selected = model.ifEmpty { null },
onSelect = { chosen -> model = if (model == chosen) "" else chosen },
)
}
}
Spacer(Modifier.height(8.dp))
OutlinedTextField(
@@ -271,7 +313,7 @@ fun SpawnScreen(
ChipGroup(
label = "Permissions",
options = PERMISSION_MODES,
options = current.permissionModes,
selected = permissionMode,
onSelect = { permissionMode = it },
)
+20
View File
@@ -258,6 +258,26 @@ impl DriverKind {
Self::Echo | Self::LlamaCpp => false,
}
}
/// Permission choices the phone can offer for this kind, in display order.
/// The driver remains responsible for translating these stable values into
/// its CLI's arguments or control protocol.
pub fn permission_modes(self) -> &'static [&'static str] {
match self {
Self::ClaudeCli => &["manual", "acceptEdits", "auto", "bypassPermissions", "plan"],
Self::CodexCli => &["workspace-write", "read-only", "danger-full-access"],
Self::Echo | Self::LlamaCpp => &[],
}
}
/// The mode used when a new-session form first selects this kind.
pub fn default_permission_mode(self) -> Option<&'static str> {
match self {
Self::ClaudeCli => Some("auto"),
Self::CodexCli => Some("workspace-write"),
Self::Echo | Self::LlamaCpp => None,
}
}
}
#[derive(Debug, Clone, Serialize, Deserialize)]
+30
View File
@@ -8,6 +8,7 @@
//! POST /setups/probe dry run {ssh?}: what would be found there
//! GET /setups/{id} one machine, for refetching after a change
//! GET /setups/{id}/models GGUFs on that machine, for a llama session
//! GET /setups/{id}/providers/{provider}/models models a CLI currently offers
//! GET /setups/{id}/dir?path=P entries of directory P, and P resolved
//! GET /setups/{id}/file?path=P content of file P, or why not
//! PUT /setups/{id}/file {path, content, ifSha256} -> new size/mtime/sha256
@@ -127,6 +128,10 @@ pub fn router(manager: Arc<SessionManager>) -> Router {
)
// The models on the machine a setup names, for a llama session there.
.route("/setups/{id}/models", get(setup_models))
.route(
"/setups/{id}/providers/{provider}/models",
get(provider_models),
)
// The filesystem of the machine a setup names. Under the setup
// rather than under a session because a filesystem is a property of
// a machine; a session only says where to start looking.
@@ -290,6 +295,9 @@ struct ProviderInfo {
name: String,
kind: crate::config::DriverKind,
models: Vec<String>,
permission_modes: Vec<&'static str>,
#[serde(skip_serializing_if = "Option::is_none")]
default_permission_mode: Option<&'static str>,
}
async fn list_setups(State(manager): State<Arc<SessionManager>>) -> axum::Json<Vec<SetupInfo>> {
@@ -308,6 +316,8 @@ fn info_for(setup: crate::config::SetupConfig) -> SetupInfo {
name: provider.name,
kind: provider.kind,
models: provider.models,
permission_modes: provider.kind.permission_modes().to_vec(),
default_permission_mode: provider.kind.default_permission_mode(),
})
.collect(),
}
@@ -412,6 +422,8 @@ async fn probe_setup(
name: provider.name,
kind: provider.kind,
models: provider.models,
permission_modes: provider.kind.permission_modes().to_vec(),
default_permission_mode: provider.kind.default_permission_mode(),
})
.collect(),
))
@@ -570,6 +582,24 @@ async fn setup_models(
.map_err(from_machine)
}
/// The models a CLI provider currently offers on the setup's machine.
/// Codex answers from its live account catalog; providers with a configured
/// shortcut list return that list.
async fn provider_models(
State(manager): State<Arc<SessionManager>>,
UrlPath((id, provider_name)): UrlPath<(String, String)>,
) -> Result<axum::Json<Vec<String>>, ApiError> {
let setup = setup_by_id(&manager, &id)?;
let provider = setup.provider(&provider_name).ok_or_else(|| {
ApiError::NotFound(format!("no provider {provider_name} on {}", setup.name))
})?;
let transport = crate::session::transport::Transport::for_setup(&setup);
crate::setups::provider_models(&transport, provider)
.await
.map(axum::Json)
.map_err(from_machine)
}
/// What is in a directory, and what that directory resolved to.
async fn list_dir(
State(manager): State<Arc<SessionManager>>,
+5 -3
View File
@@ -303,14 +303,16 @@ fn start_process(
let settings = inner.settings.lock().unwrap();
let mut args = Vec::new();
match settings.permission_mode.as_deref() {
Some("bypassPermissions") => {
Some("bypassPermissions" | "danger-full-access") => {
args.push("--dangerously-bypass-approvals-and-sandbox".to_string());
}
Some("manual") => {
args.extend(["--ask-for-approval".to_string(), "on-request".to_string()]);
}
Some("auto" | "acceptEdits") => args.push("--approve-for-me".to_string()),
Some("plan") => args.extend([
Some("auto" | "acceptEdits" | "workspace-write") => {
args.push("--approve-for-me".to_string());
}
Some("plan" | "read-only") => args.extend([
"--ask-for-approval".to_string(),
"never".to_string(),
"--sandbox".to_string(),
+1
View File
@@ -91,6 +91,7 @@ pub enum Streams {
}
/// The machine a session's process runs on.
#[derive(Clone)]
pub enum Transport {
/// The machine this server is running on.
Here,
+78 -1
View File
@@ -14,7 +14,8 @@
//! is editing `config.ron` on the backend, which is exactly the authority the
//! phone is not being given.
use anyhow::Result;
use anyhow::{Context, Result};
use serde_json::{Value, json};
use crate::config::{DriverKind, ProviderConfig};
use crate::session::transport::{Launch, Transport};
@@ -87,6 +88,60 @@ pub async fn discover(transport: &Transport) -> Result<Vec<ProviderConfig>> {
Ok(providers)
}
/// Models the selected provider currently offers on this machine.
///
/// Codex's catalog is account- and CLI-version-specific, so it is asked at the
/// moment the picker opens rather than copied into `config.ron`. Other
/// providers retain the shortcut list discovery stored for them.
pub async fn provider_models(
transport: &Transport,
provider: &ProviderConfig,
) -> Result<Vec<String>> {
if provider.kind != DriverKind::CodexCli {
return Ok(provider.models.clone());
}
let transport = transport.clone();
let program = provider.program().to_string();
tokio::task::spawn_blocking(move || {
let launch = Launch::new(program, vec!["app-server".into(), "--stdio".into()], None);
let initial = json!({
"id": 1,
"method": "initialize",
"params": {"clientInfo": {"name": "ai-app", "title": "AI Sessions", "version": env!("CARGO_PKG_VERSION")}}
});
let requests = [
json!({"method": "initialized"}),
json!({"id": 2, "method": "model/list", "params": {"includeHidden": false, "limit": 100}}),
];
let answer = transport.request_json_blocking(&launch, &initial, &requests, 2)?;
parse_codex_models(&answer)
})
.await?
}
fn parse_codex_models(answer: &Value) -> Result<Vec<String>> {
if let Some(message) = answer.pointer("/error/message").and_then(Value::as_str) {
anyhow::bail!("Codex could not list models: {message}");
}
let entries = answer
.pointer("/result/data")
.and_then(Value::as_array)
.context("Codex returned no model catalog")?;
let mut models = entries
.iter()
.filter(|entry| {
!entry
.get("hidden")
.and_then(Value::as_bool)
.unwrap_or(false)
})
.filter_map(|entry| entry.get("model").and_then(Value::as_str))
.map(str::to_string)
.collect::<Vec<_>>();
models.dedup();
Ok(models)
}
/// Adds what to do to failures whose own wording does not say.
///
/// ssh's messages are written for someone at a terminal on the backend, which is
@@ -213,4 +268,26 @@ mod tests {
assert_eq!(shorten_home(&sibling), sibling);
assert_eq!(shorten_home("/etc/hosts"), "/etc/hosts");
}
#[test]
fn codex_models_are_the_selectable_non_hidden_catalog_entries() {
let answer = json!({"result": {"data": [
{"model": "gpt-small", "hidden": false},
{"model": "gpt-hidden", "hidden": true},
{"model": "gpt-large", "hidden": false}
]}});
assert_eq!(
parse_codex_models(&answer).unwrap(),
vec!["gpt-small".to_string(), "gpt-large".to_string()]
);
}
#[test]
fn a_failed_codex_catalog_is_not_reported_as_an_empty_one() {
let answer = json!({"error": {"message": "login required"}});
assert_eq!(
parse_codex_models(&answer).unwrap_err().to_string(),
"Codex could not list models: login required"
);
}
}