Add local CLI access via Unix-socket server

New CLIServerService listens on a Unix domain socket
(~/Library/Application Support/oAI/cli.sock) speaking a minimal
HTTP/1.1 subset, for one-shot non-streaming shell access to a single
fixed model — e.g. an `ai "prompt"` zsh function — without opening
the app window and without going through the tool-calling loop.

Configured in Settings > MCP > CLI Access (toggle, provider, model —
deliberately independent of the chat UI's active model). JSON
request/response envelope rather than raw text so new fields (model
override, streaming, tool support) can be added later without a
breaking wire-format change.

Verified live end-to-end against a real OpenRouter request, error
paths, and clean-shutdown socket cleanup. 10 new unit tests cover
the HTTP framing/parsing logic.
This commit is contained in:
2026-08-05 13:33:02 +02:00
parent 2adce758f1
commit 897bfdce10
6 changed files with 552 additions and 0 deletions
+247
View File
@@ -0,0 +1,247 @@
//
// CLIServerService.swift
// Confab
//
// Local Unix-socket server for one-shot shell/CLI access to a single fixed model
//
// SPDX-License-Identifier: PolyForm-Noncommercial-1.0.0
// Copyright (C) 2026 Rune Olsen
//
// This file is part of Confab.
//
// Confab is licensed under the PolyForm Noncommercial License 1.0.0.
// You may use, study, modify, and share it for any noncommercial
// purpose. Commercial use including selling Confab or any part of
// it, standalone or bundled into another product or service
// requires a separate commercial license from the copyright holder.
//
// See the LICENSE file or
// <https://polyformproject.org/licenses/noncommercial/1.0.0> for
// the full license text. For commercial licensing, contact Rune
// Olsen via <https://confab.no>.
import Foundation
import Network
import os
/// Local Unix-domain-socket server exposing one-shot, non-streaming text completions from a
/// single fixed model (Settings MCP CLI Access) e.g. an `ai "prompt"` shell function.
/// Deliberately minimal for v1: no tool-calling, no streaming, no per-request model override.
/// Speaks a minimal HTTP/1.1 subset (just enough for `curl --unix-socket`) rather than a custom
/// protocol, so it stays curl-friendly and trivially extensible new fields can be added to the
/// JSON request/response bodies later without changing the transport.
final class CLIServerService {
static let shared = CLIServerService()
private init() {}
private let log = Logger(subsystem: Log.subsystem, category: "cli")
/// All listener/connection callbacks run on this single serial queue, so `activeConnections`
/// never needs a lock Network.framework callbacks don't run on the main thread.
private let queue = DispatchQueue(label: "com.oai.Confab.cliserver")
private var listener: NWListener?
private var activeConnections: [ObjectIdentifier: NWConnection] = [:]
static let socketPath: String = {
(("~/Library/Application Support/oAI" as NSString).expandingTildeInPath as NSString)
.appendingPathComponent("cli.sock")
}()
// MARK: - Lifecycle
func start() {
guard SettingsService.shared.cliServerEnabled else {
log.info("CLI server disabled")
return
}
startListening()
}
func stop() {
listener?.cancel()
listener = nil
for (_, connection) in activeConnections { connection.cancel() }
activeConnections.removeAll()
try? FileManager.default.removeItem(atPath: Self.socketPath)
log.info("CLI server stopped")
}
/// Call after toggling cliServerEnabled or changing the provider/model, so the change takes
/// effect immediately instead of requiring an app relaunch.
func restart() {
stop()
start()
}
private func startListening() {
// Remove a stale socket file left behind by an unclean shutdown (bind() fails on an
// existing path even if nothing is listening on it anymore).
try? FileManager.default.removeItem(atPath: Self.socketPath)
let params = NWParameters()
params.defaultProtocolStack.transportProtocol = NWProtocolTCP.Options()
params.requiredLocalEndpoint = NWEndpoint.unix(path: Self.socketPath)
params.allowLocalEndpointReuse = true
do {
let newListener = try NWListener(using: params)
newListener.newConnectionHandler = { [weak self] connection in
self?.handle(connection: connection)
}
newListener.stateUpdateHandler = { [weak self] state in
switch state {
case .ready:
self?.log.info("CLI server listening at \(Self.socketPath)")
case .failed(let error):
self?.log.error("CLI server listener failed: \(error.localizedDescription)")
default:
break
}
}
newListener.start(queue: queue)
listener = newListener
} catch {
log.error("Failed to start CLI server: \(error.localizedDescription)")
}
}
// MARK: - Connection handling
private func handle(connection: NWConnection) {
let id = ObjectIdentifier(connection)
activeConnections[id] = connection
connection.stateUpdateHandler = { [weak self] state in
switch state {
case .failed, .cancelled:
self?.activeConnections.removeValue(forKey: id)
default:
break
}
}
connection.start(queue: queue)
readRequest(on: connection, buffer: Data())
}
private func readRequest(on connection: NWConnection, buffer: Data) {
connection.receive(minimumIncompleteLength: 1, maximumLength: 65536) { [weak self] data, _, isComplete, error in
guard let self else { return }
var buffer = buffer
if let data { buffer.append(data) }
if let body = Self.parseRequestBody(from: buffer) {
Task {
let responseData = await self.processRequest(body)
self.sendAndClose(responseData, on: connection)
}
return
}
if isComplete || error != nil {
connection.cancel()
return
}
self.readRequest(on: connection, buffer: buffer)
}
}
private func sendAndClose(_ data: Data, on connection: NWConnection) {
let id = ObjectIdentifier(connection)
connection.send(content: data, completion: .contentProcessed { [weak self] _ in
connection.cancel()
self?.activeConnections.removeValue(forKey: id)
})
}
// MARK: - Request processing
private struct AskRequest: Decodable {
let prompt: String
}
private func processRequest(_ body: Data) async -> Data {
guard let req = try? JSONDecoder().decode(AskRequest.self, from: body),
!req.prompt.trimmingCharacters(in: .whitespacesAndNewlines).isEmpty else {
return Self.errorResponse("Request body must be JSON: {\"prompt\": \"...\"}")
}
let settings = SettingsService.shared
guard settings.cliServerEnabled else {
return Self.errorResponse("CLI server is disabled in Settings")
}
guard let providerType = Settings.Provider(rawValue: settings.cliServerProvider),
let provider = ProviderRegistry.shared.getProvider(for: providerType) else {
return Self.errorResponse("No provider configured — set one in Settings > MCP > CLI Access")
}
guard !settings.cliServerModel.isEmpty else {
return Self.errorResponse("No model configured — set one in Settings > MCP > CLI Access")
}
do {
let request = ChatRequest(
messages: [Message(role: .user, content: req.prompt)],
model: settings.cliServerModel,
stream: false
)
let response = try await provider.chat(request: request)
return Self.successResponse(response.content)
} catch {
log.error("CLI request failed: \(error.localizedDescription)")
return Self.errorResponse(error.localizedDescription, statusCode: 500)
}
}
// MARK: - Pure helpers (unit tested see CLIServerServiceTests.swift)
struct AskResponseBody: Codable {
let response: String?
let error: String?
}
/// Parses a minimal HTTP/1.1 request out of `buffer`, returning the body once the full
/// header block and (per Content-Length) body have arrived, or nil if more data is needed.
/// Not a general-purpose HTTP parser this server only ever talks to `curl --unix-socket`,
/// so the method/path/most headers are read past rather than validated.
static func parseRequestBody(from buffer: Data) -> Data? {
let headerTerminator = Data("\r\n\r\n".utf8)
guard let headerEndRange = buffer.range(of: headerTerminator) else { return nil }
let headerData = buffer.subdata(in: buffer.startIndex..<headerEndRange.lowerBound)
guard let headerString = String(data: headerData, encoding: .utf8) else { return nil }
var contentLength = 0
for line in headerString.split(separator: "\r\n") {
let parts = line.split(separator: ":", maxSplits: 1)
guard parts.count == 2,
parts[0].trimmingCharacters(in: .whitespaces).caseInsensitiveCompare("Content-Length") == .orderedSame,
let value = Int(parts[1].trimmingCharacters(in: .whitespaces)) else { continue }
contentLength = value
}
let bodyStart = headerEndRange.upperBound
let availableBodyLength = buffer.distance(from: bodyStart, to: buffer.endIndex)
guard availableBodyLength >= contentLength else { return nil }
let bodyEnd = buffer.index(bodyStart, offsetBy: contentLength)
return buffer.subdata(in: bodyStart..<bodyEnd)
}
static func makeHTTPResponse(statusCode: Int, statusText: String, jsonBody: Data) -> Data {
let header = "HTTP/1.1 \(statusCode) \(statusText)\r\nContent-Type: application/json\r\nContent-Length: \(jsonBody.count)\r\nConnection: close\r\n\r\n"
var data = Data(header.utf8)
data.append(jsonBody)
return data
}
static func successResponse(_ text: String) -> Data {
let body = (try? JSONEncoder().encode(AskResponseBody(response: text, error: nil))) ?? Data()
return makeHTTPResponse(statusCode: 200, statusText: "OK", jsonBody: body)
}
static func errorResponse(_ message: String, statusCode: Int = 400) -> Data {
let body = (try? JSONEncoder().encode(AskResponseBody(response: nil, error: message))) ?? Data()
let statusText = statusCode == 400 ? "Bad Request" : "Internal Server Error"
return makeHTTPResponse(statusCode: statusCode, statusText: statusText, jsonBody: body)
}
}