yanis/jan
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Merge branch 'main' of https://github.com/janhq/jan into feat/assistant-chat-button

Browse Source
This commit is contained in:
markmehere
2026-03-24 09:17:45 +11:00
parent 2a80d10a51 080f8fc7f7
commit 344d830e97
27 changed files with 815 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
.gitignore vendored
View File

@@ -55,6 +55,11 @@ docs/.next/
**/yarn-error.log*
**/pnpm-debug.log*
## Swift Package Manager (Foundation Models server)
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/.build/
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/.swiftpm/
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/Package.resolved
## cargo
target
Cargo.lock

44
Makefile
View File

@@ -6,15 +6,28 @@ REPORT_PORTAL_PROJECT_NAME ?= ""
REPORT_PORTAL_LAUNCH_NAME ?= "Jan App"
REPORT_PORTAL_DESCRIPTION ?= "Jan App report"
# Detect OS
ifeq ($(OS),Windows_NT)
DETECTED_OS := Windows
else
DETECTED_OS := $(shell uname -s)
endif
ifeq ($(OS),Windows_NT)
MKDIR = if not exist "$(1)" mkdir "$(1)"
else
MKDIR = mkdir -p $(1)
endif
# Default target, does nothing
all:
@echo "Specify a target to run"
# Installs yarn dependencies and builds core and extensions
install-and-build:
ifeq ($(OS),Windows_NT)
ifeq ($(DETECTED_OS),Windows)
echo "skip"
else ifeq ($(shell uname -s),Linux)
else ifeq ($(DETECTED_OS),Linux)
chmod +x src-tauri/build-utils/*
endif
yarn install
@@ -24,7 +37,7 @@ endif
# Install required Rust targets for macOS universal builds
install-rust-targets:
ifeq ($(shell uname -s),Darwin)
ifeq ($(DETECTED_OS),Darwin)
@echo "Detected macOS, installing universal build targets..."
rustup target add x86_64-apple-darwin
rustup target add aarch64-apple-darwin
@@ -88,7 +101,7 @@ dev-android: install-and-build install-android-rust-targets
dev-ios: install-and-build install-ios-rust-targets
@echo "Setting up iOS development environment..."
ifeq ($(shell uname -s),Darwin)
ifeq ($(DETECTED_OS),Darwin)
@if [ ! -d "src-tauri/gen/ios" ]; then \
echo "iOS app not initialized. Initializing..."; \
yarn tauri ios init; \
@@ -110,7 +123,7 @@ lint: install-and-build
# Testing
test: lint install-rust-targets
yarn download:bin
ifeq ($(OS),Windows_NT)
ifeq ($(DETECTED_OS),Windows)
endif
yarn test
yarn copy:assets:tauri
@@ -124,7 +137,7 @@ endif
# Build MLX server (macOS Apple Silicon only) - always builds
build-mlx-server:
ifeq ($(shell uname -s),Darwin)
ifeq ($(DETECTED_OS),Darwin)
@echo "Building MLX server for Apple Silicon..."
cd mlx-server && swift build -c release
@echo "Copying build products..."
@@ -162,7 +175,7 @@ endif
# Build MLX server only if not already present (for dev)
build-mlx-server-if-exists:
ifeq ($(shell uname -s),Darwin)
ifeq ($(DETECTED_OS),Darwin)
@if [ -f "src-tauri/resources/bin/mlx-server" ]; then \
echo "MLX server already exists at src-tauri/resources/bin/mlx-server, skipping build..."; \
else \
@@ -172,10 +185,9 @@ else
@echo "Skipping MLX server build (macOS only)"
endif
# Build jan CLI (release, platform-aware) → src-tauri/resources/bin/jan[.exe]
build-cli:
ifeq ($(shell uname -s),Darwin)
ifeq ($(DETECTED_OS),Darwin)
cd src-tauri && cargo build --release --features cli --bin jan-cli --target aarch64-apple-darwin
cd src-tauri && cargo build --release --features cli --bin jan-cli --target x86_64-apple-darwin
lipo -create \
@@ -183,7 +195,7 @@ ifeq ($(shell uname -s),Darwin)
src-tauri/target/x86_64-apple-darwin/release/jan-cli \
-output src-tauri/resources/bin/jan-cli
chmod +x src-tauri/resources/bin/jan-cli
mkdir -p src-tauri/target/universal-apple-darwin/release
$(call MKDIR,'src-tauri/target/universal-apple-darwin/release')
echo "Checking for code signing identity..."; \
SIGNING_IDENTITY=$$(security find-identity -v -p codesigning | grep "Developer ID Application" | head -1 | sed 's/.*"\(.*\)".*/\1/'); \
@@ -196,7 +208,7 @@ ifeq ($(shell uname -s),Darwin)
fi
cp src-tauri/resources/bin/jan-cli src-tauri/target/universal-apple-darwin/release/jan-cli
else ifeq ($(OS),Windows_NT)
else ifeq ($(DETECTED_OS),Windows)
cd src-tauri && cargo build --release --features cli --bin jan-cli
cp src-tauri/target/release/jan-cli.exe src-tauri/resources/bin/jan-cli.exe
else
@@ -206,16 +218,20 @@ endif
# Debug build for local dev (faster, native arch only)
build-cli-dev:
mkdir -p src-tauri/resources/bin
$(call MKDIR,'src-tauri/resources/bin')
cd src-tauri && cargo build --features cli --bin jan-cli
ifeq ($(DETECTED_OS),Windows)
copy src-tauri\target\debug\jan-cli.exe src-tauri\resources\bin\jan-cli.exe
else
install -m755 src-tauri/target/debug/jan-cli src-tauri/resources/bin/jan-cli
endif
# Build
build: install-and-build install-rust-targets
yarn build
clean:
ifeq ($(OS),Windows_NT)
ifeq ($(DETECTED_OS),Windows)
-powershell -Command "Get-ChildItem -Path . -Include node_modules, .next, dist, build, out, .turbo, .yarn -Recurse -Directory | Remove-Item -Recurse -Force"
-powershell -Command "Get-ChildItem -Path . -Include package-lock.json, tsconfig.tsbuildinfo -Recurse -File | Remove-Item -Recurse -Force"
-powershell -Command "Remove-Item -Recurse -Force ./pre-install/*.tgz"
@@ -224,7 +240,7 @@ ifeq ($(OS),Windows_NT)
-powershell -Command "Remove-Item -Recurse -Force ./src-tauri/resources"
-powershell -Command "Remove-Item -Recurse -Force ./src-tauri/target"
-powershell -Command "if (Test-Path \"$($env:USERPROFILE)\jan\extensions\") { Remove-Item -Path \"$($env:USERPROFILE)\jan\extensions\" -Recurse -Force }"
else ifeq ($(shell uname -s),Linux)
else ifeq ($(DETECTED_OS),Linux)
find . -name "node_modules" -type d -prune -exec rm -rf '{}' +
find . -name ".next" -type d -exec rm -rf '{}' +
find . -name "dist" -type d -exec rm -rf '{}' +

33
docs/src/pages/changelog/2026-03-23-jan-remote-models-context-cap.mdx Normal file
View File

@@ -0,0 +1,33 @@
---
title: "Jan v0.7.9: CLI on Windows, Smarter Context Management & Safer Data Location Management"
version: 0.7.9
description: "Jan v0.7.9 fetches the latest models during onboarding, caps context length to avoid high RAM usage, fixes CLI on Windows, and improves data location management."
date: 2026-03-23
---
import ChangelogHeader from "@/components/Changelog/ChangelogHeader"
import { Callout } from 'nextra/components'
<ChangelogHeader
title="Jan v0.7.9: CLI on Windows, Smarter Context Management & Safer Data Location Management"
date="2026-03-23"
/>
# Highlights
**Remove OpenClaw Integration**
We recognize that the OpenClaw integration has been causing issues for some users. OpenClaw is a popular agent that has received significant attention from the community. However, it is a resource-heavy product that consumes a large amount of memory and tokens, and can introduce critical security vulnerabilities if not carefully handled in a local AI environment. Therefore, we have decided to remove OpenClaw from Jan and shift our focus toward exploring and building a simpler, safer, and more practical agent experience for Jan users in the near future.
- Jan now remotely fetches the latest models during onboarding, ready for upcoming new Jan models
- Context length is now automatically capped and increased reasonably, avoiding excessively large context sizes that caused high RAM usage
- Fixed CLI path installation on Windows
- Fixed changing Jan data location
- Safely remove Jan data folder with a registered list
---
Update your Jan or [download the latest](https://jan.ai/).
For the complete list of changes, see the [GitHub release notes](https://github.com/janhq/jan/releases/tag/v0.7.9)

13
extensions/llamacpp-extension/src/index.ts
View File

@@ -64,7 +64,6 @@ import {
import { getSystemUsage, getSystemInfo } from '@janhq/tauri-plugin-hardware-api'
// Error message constant - matches web-app/src/utils/error.ts
const OUT_OF_CONTEXT_SIZE = 'the request exceeds the available context size.'
/**
* Override the default app.log function to use Jan's logging system.
@@ -1830,12 +1829,6 @@ export default class llamacpp_extension extends AIEngine {
const data = JSON.parse(jsonStr)
const chunk = data as chatCompletionChunk
// Check for out-of-context error conditions
if (chunk.choices?.[0]?.finish_reason === 'length') {
// finish_reason 'length' indicates context limit was hit
throw new Error(OUT_OF_CONTEXT_SIZE)
}
yield chunk
} catch (e) {
logger.error('Error parsing JSON from stream or server error:', e)
@@ -1923,12 +1916,6 @@ export default class llamacpp_extension extends AIEngine {
const completionResponse = (await response.json()) as chatCompletion
// Check for out-of-context error conditions
if (completionResponse.choices?.[0]?.finish_reason === 'length') {
// finish_reason 'length' indicates context limit was hit
throw new Error(OUT_OF_CONTEXT_SIZE)
}
return completionResponse
}

34
extensions/yarn.lock
View File

@@ -376,6 +376,18 @@ __metadata:
languageName: node
linkType: hard
"@janhq/core@file:../../core/package.tgz::locator=%40janhq%2Ffoundation-models-extension%40workspace%3Afoundation-models-extension":
version: 0.1.10
resolution: "@janhq/core@file:../../core/package.tgz#../../core/package.tgz::hash=6c8a42&locator=%40janhq%2Ffoundation-models-extension%40workspace%3Afoundation-models-extension"
dependencies:
rxjs: "npm:^7.8.1"
ulidx: "npm:^2.3.0"
peerDependencies:
react: 19.0.0
checksum: 10c0/c80736877d9b0d9498d76588a1e92664f467a84265aee7cb6be497b89ed4b1e1c9379c12686c4676bb438eb268ba97caae607b01686e3dc0d485933a4ab69af7
languageName: node
linkType: hard
"@janhq/core@file:../../core/package.tgz::locator=%40janhq%2Fllamacpp-extension%40workspace%3Allamacpp-extension":
version: 0.1.10
resolution: "@janhq/core@file:../../core/package.tgz#../../core/package.tgz::hash=6c8a42&locator=%40janhq%2Fllamacpp-extension%40workspace%3Allamacpp-extension"
@@ -439,6 +451,22 @@ __metadata:
languageName: unknown
linkType: soft
"@janhq/foundation-models-extension@workspace:foundation-models-extension":
version: 0.0.0-use.local
resolution: "@janhq/foundation-models-extension@workspace:foundation-models-extension"
dependencies:
"@janhq/core": ../../core/package.tgz
"@janhq/tauri-plugin-foundation-models-api": "link:../../src-tauri/plugins/tauri-plugin-foundation-models"
"@tauri-apps/api": "npm:2.8.0"
"@tauri-apps/plugin-http": "npm:2.5.0"
"@tauri-apps/plugin-log": "npm:^2.6.0"
cpx: "npm:1.5.0"
rimraf: "npm:3.0.2"
rolldown: "npm:1.0.0-beta.1"
typescript: "npm:5.9.2"
languageName: unknown
linkType: soft
"@janhq/llamacpp-extension@workspace:llamacpp-extension":
version: 0.0.0-use.local
resolution: "@janhq/llamacpp-extension@workspace:llamacpp-extension"
@@ -493,6 +521,12 @@ __metadata:
languageName: unknown
linkType: soft
"@janhq/tauri-plugin-foundation-models-api@link:../../src-tauri/plugins/tauri-plugin-foundation-models::locator=%40janhq%2Ffoundation-models-extension%40workspace%3Afoundation-models-extension":
version: 0.0.0-use.local
resolution: "@janhq/tauri-plugin-foundation-models-api@link:../../src-tauri/plugins/tauri-plugin-foundation-models::locator=%40janhq%2Ffoundation-models-extension%40workspace%3Afoundation-models-extension"
languageName: node
linkType: soft
"@janhq/tauri-plugin-hardware-api@link:../../src-tauri/plugins/tauri-plugin-hardware::locator=%40janhq%2Fllamacpp-extension%40workspace%3Allamacpp-extension":
version: 0.0.0-use.local
resolution: "@janhq/tauri-plugin-hardware-api@link:../../src-tauri/plugins/tauri-plugin-hardware::locator=%40janhq%2Fllamacpp-extension%40workspace%3Allamacpp-extension"

30
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/Package.swift Normal file
View File

@@ -0,0 +1,30 @@
// swift-tools-version: 6.2
import PackageDescription
let package = Package(
name: "foundation-models-server",
platforms: [
.macOS(.v26)
],
products: [
.executable(name: "foundation-models-server", targets: ["FoundationModelsServer"])
],
dependencies: [
.package(url: "https://github.com/apple/swift-argument-parser", from: "1.7.0"),
.package(url: "https://github.com/hummingbird-project/hummingbird", from: "2.19.0"),
],
targets: [
.executableTarget(
name: "FoundationModelsServer",
dependencies: [
.product(name: "ArgumentParser", package: "swift-argument-parser"),
.product(name: "Hummingbird", package: "hummingbird"),
],
path: "Sources/FoundationModelsServer",
swiftSettings: [
.swiftLanguageMode(.v6)
]
)
]
)

40
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/README.md Normal file
View File

@@ -0,0 +1,40 @@
# foundation-models-server
A lightweight OpenAI-compatible HTTP server that wraps Apple's Foundation Models framework, enabling Jan to use on-device Apple Intelligence models on macOS 26+.
## Requirements
- macOS 26 (Tahoe) or later
- Apple Silicon Mac with Apple Intelligence enabled
- Xcode 26 or later
## Building
```bash
swift build -c release
```
The binary will be at `.build/release/foundation-models-server`.
## Usage
```bash
# Check availability
foundation-models-server --check
# Start server on default port
foundation-models-server --port 8080
# Start server with API key
foundation-models-server --port 8080 --api-key <key>
```
## API
The server exposes an OpenAI-compatible API:
- `GET /health` — health check
- `GET /v1/models` — lists the `apple/on-device` model
- `POST /v1/chat/completions` — chat completions (streaming and non-streaming)
The model ID is always `apple/on-device`.

78
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/Sources/FoundationModelsServer/FoundationModelsServerCommand.swift Normal file
View File

@@ -0,0 +1,78 @@
import ArgumentParser
import Foundation
import Hummingbird
import FoundationModels
@main
struct FoundationModelsServerCommand: AsyncParsableCommand {
static let configuration = CommandConfiguration(
commandName: "foundation-models-server",
abstract: "Apple Foundation Models inference server with OpenAI-compatible API"
)
@Option(name: .long, help: "Port to listen on")
var port: Int = 8080
@Option(name: .long, help: "API key for authentication (optional)")
var apiKey: String = ""
@Flag(name: .long, help: "Check availability and exit with status 0 if available")
var check: Bool = false
func run() async throws {
let availability = SystemLanguageModel.default.availability
// In --check mode, always print a machine-readable status token and exit 0.
// Callers (e.g. the Tauri plugin) parse this string to decide visibility.
if check {
switch availability {
case .available:
print("available")
case .unavailable(.deviceNotEligible):
print("notEligible")
case .unavailable(.appleIntelligenceNotEnabled):
print("appleIntelligenceNotEnabled")
case .unavailable(.modelNotReady):
print("modelNotReady")
default:
print("unavailable")
}
return
}
guard case .available = availability else {
let reason: String
switch availability {
case .unavailable(.deviceNotEligible):
reason = "Device is not eligible for Apple Intelligence"
case .unavailable(.appleIntelligenceNotEnabled):
reason = "Apple Intelligence is not enabled in System Settings"
case .unavailable(.modelNotReady):
reason = "Foundation model is downloading or not yet ready"
default:
reason = "Foundation model is unavailable on this system"
}
fputs("[foundation-models] ERROR: \(reason)\n", stderr)
throw ExitCode(1)
}
log("[foundation-models] Foundation Models Server starting...")
log("[foundation-models] Port: \(port)")
let server = FoundationModelsHTTPServer(
modelId: "apple/on-device",
apiKey: apiKey
)
let router = server.buildRouter()
let app = Application(
router: router,
configuration: .init(address: .hostname("127.0.0.1", port: port))
)
log("[foundation-models] http server listening on http://127.0.0.1:\(port)")
log("[foundation-models] server is listening on 127.0.0.1:\(port)")
try await app.run()
}
}

6
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/Sources/FoundationModelsServer/Logger.swift Normal file
View File

@@ -0,0 +1,6 @@
import Foundation
func log(_ message: String) {
print(message)
fflush(stdout)
}

299
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/Sources/FoundationModelsServer/Server.swift Normal file
View File

@@ -0,0 +1,299 @@
import Foundation
import Hummingbird
import FoundationModels
/// HTTP server exposing an OpenAI-compatible API backed by Apple Foundation Models
struct FoundationModelsHTTPServer: Sendable {
let modelId: String
let apiKey: String
func buildRouter() -> Router<BasicRequestContext> {
let router = Router()
// Health check
router.get("/health") { _, _ in
let response = HealthResponse(status: "ok")
return try encodeJSONResponse(response)
}
// List available models
router.get("/v1/models") { _, _ in
let response = ModelsListResponse(
object: "list",
data: [
ModelData(
id: self.modelId,
object: "model",
created: currentTimestamp(),
owned_by: "apple"
)
]
)
return try encodeJSONResponse(response)
}
// Chat completions (OpenAI-compatible)
router.post("/v1/chat/completions") { request, _ in
// Validate API key when configured
if !self.apiKey.isEmpty {
let authHeader = request.headers[.authorization]
guard authHeader == "Bearer \(self.apiKey)" else {
let errorResp = ErrorResponse(
error: ErrorDetail(
message: "Unauthorized: invalid or missing API key",
type: "authentication_error",
code: "unauthorized"
)
)
return try Response(
status: .unauthorized,
headers: [.contentType: "application/json"],
body: .init(byteBuffer: encodeJSONBuffer(errorResp))
)
}
}
let body = try await request.body.collect(upTo: 10 * 1024 * 1024)
let chatRequest: ChatCompletionRequest
do {
chatRequest = try JSONDecoder().decode(ChatCompletionRequest.self, from: body)
} catch {
let errorResp = ErrorResponse(
error: ErrorDetail(
message: "Invalid request body: \(error.localizedDescription)",
type: "invalid_request_error",
code: nil
)
)
return try Response(
status: .badRequest,
headers: [.contentType: "application/json"],
body: .init(byteBuffer: encodeJSONBuffer(errorResp))
)
}
let isStreaming = chatRequest.stream ?? false
log("[foundation-models] Request: messages=\(chatRequest.messages.count), stream=\(isStreaming)")
if isStreaming {
return try await self.handleStreamingRequest(chatRequest)
} else {
return try await self.handleNonStreamingRequest(chatRequest)
}
}
return router
}
// MARK: - Non-streaming
private func handleNonStreamingRequest(_ chatRequest: ChatCompletionRequest) async throws -> Response {
let session = buildSession(from: chatRequest.messages)
let lastUserMessage = extractLastUserMessage(from: chatRequest.messages)
let response = try await session.respond(to: lastUserMessage)
let content = response.content
let completionResponse = ChatCompletionResponse(
id: "chatcmpl-\(UUID().uuidString)",
object: "chat.completion",
created: currentTimestamp(),
model: modelId,
choices: [
ChatCompletionChoice(
index: 0,
message: ChatResponseMessage(role: "assistant", content: content),
finish_reason: "stop"
)
],
usage: UsageInfo(
prompt_tokens: 0,
completion_tokens: 0,
total_tokens: 0
)
)
return try encodeJSONResponse(completionResponse)
}
// MARK: - Streaming
private func handleStreamingRequest(_ chatRequest: ChatCompletionRequest) async throws -> Response {
let requestId = "chatcmpl-\(UUID().uuidString)"
let created = currentTimestamp()
let modelId = self.modelId
let messages = chatRequest.messages
let (stream, continuation) = AsyncStream<ByteBuffer>.makeStream()
let task = Task { [self] in
do {
let session = self.buildSession(from: messages)
let lastUserMessage = self.extractLastUserMessage(from: messages)
let roleDelta = ChatCompletionChunk(
id: requestId,
object: "chat.completion.chunk",
created: created,
model: modelId,
choices: [
ChunkChoice(
index: 0,
delta: DeltaContent(role: "assistant", content: nil),
finish_reason: nil
)
]
)
if let buffer = encodeSSEBuffer(roleDelta) {
continuation.yield(buffer)
}
var previousText = ""
for try await snapshot in session.streamResponse(to: lastUserMessage) {
let currentText = snapshot.content
let delta = String(currentText.dropFirst(previousText.count))
previousText = currentText
if delta.isEmpty { continue }
let chunk = ChatCompletionChunk(
id: requestId,
object: "chat.completion.chunk",
created: created,
model: modelId,
choices: [
ChunkChoice(
index: 0,
delta: DeltaContent(role: nil, content: delta),
finish_reason: nil
)
]
)
if let buffer = encodeSSEBuffer(chunk) {
continuation.yield(buffer)
}
}
// Send stop chunk
let stopChunk = ChatCompletionChunk(
id: requestId,
object: "chat.completion.chunk",
created: created,
model: modelId,
choices: [
ChunkChoice(
index: 0,
delta: DeltaContent(role: nil, content: nil),
finish_reason: "stop"
)
]
)
if let buffer = encodeSSEBuffer(stopChunk) {
continuation.yield(buffer)
}
// SSE terminator
var doneBuffer = ByteBufferAllocator().buffer(capacity: 16)
doneBuffer.writeString("data: [DONE]\n\n")
continuation.yield(doneBuffer)
} catch {
log("[foundation-models] Streaming error: \(error.localizedDescription)")
var errBuffer = ByteBufferAllocator().buffer(capacity: 256)
errBuffer.writeString("error: {\"message\":\"\(error.localizedDescription)\"}\n\n")
continuation.yield(errBuffer)
}
continuation.finish()
}
// Cancel the generation task when the client disconnects
continuation.onTermination = { @Sendable _ in
log("[foundation-models] SSE continuation terminated by client disconnect")
task.cancel()
}
return Response(
status: .ok,
headers: [
.contentType: "text/event-stream",
.cacheControl: "no-cache",
.init("X-Accel-Buffering")!: "no"
],
body: .init(asyncSequence: stream)
)
}
// MARK: - Session Construction
/// Build a `LanguageModelSession` from the OpenAI message list.
///
/// System messages become the session instructions.
/// Prior user/assistant turns are serialised into the instructions block so
/// the model has full conversation context without re-running inference.
/// (The Foundation Models `Transcript` API is not used for history injection
/// because it is designed for observing an already-live session's state, not
/// for priming a fresh one with arbitrary history.)
private func buildSession(from messages: [ChatMessage]) -> LanguageModelSession {
let systemContent = messages.first(where: { $0.role == "system" })?.content ?? ""
let nonSystem = messages.filter { $0.role != "system" }
let history = nonSystem.dropLast() // all turns except the last user message
var instructionsText: String
if systemContent.isEmpty {
instructionsText = "You are a helpful assistant."
} else {
instructionsText = systemContent
}
// Append prior turns so the model understands conversation context
if !history.isEmpty {
instructionsText += "\n\n[Previous conversation]\n"
for msg in history {
let label = msg.role == "assistant" ? "Assistant" : "User"
instructionsText += "\(label): \(msg.content ?? "")\n"
}
instructionsText += "[End of previous conversation]"
}
return LanguageModelSession(instructions: instructionsText)
}
private func extractLastUserMessage(from messages: [ChatMessage]) -> String {
let nonSystem = messages.filter { $0.role != "system" }
return nonSystem.last?.content ?? ""
}
}
// MARK: - Helpers
private func currentTimestamp() -> Int {
Int(Date().timeIntervalSince1970)
}
private func encodeJSONResponse<T: Encodable>(_ value: T) throws -> Response {
let data = try JSONEncoder().encode(value)
var buffer = ByteBufferAllocator().buffer(capacity: data.count)
buffer.writeBytes(data)
return Response(
status: .ok,
headers: [.contentType: "application/json"],
body: .init(byteBuffer: buffer)
)
}
private func encodeJSONBuffer<T: Encodable>(_ value: T) -> ByteBuffer {
let data = (try? JSONEncoder().encode(value)) ?? Data()
var buffer = ByteBufferAllocator().buffer(capacity: data.count)
buffer.writeBytes(data)
return buffer
}
private func encodeSSEBuffer<T: Encodable>(_ value: T) -> ByteBuffer? {
guard let json = try? JSONEncoder().encode(value),
let jsonString = String(data: json, encoding: .utf8) else {
return nil
}
let line = "data: \(jsonString)\n\n"
var buffer = ByteBufferAllocator().buffer(capacity: line.utf8.count)
buffer.writeString(line)
return buffer
}

98
src-tauri/plugins/tauri-plugin-foundation-models/swift-server/Sources/FoundationModelsServer/Types.swift Normal file
View File

@@ -0,0 +1,98 @@
import Foundation
// MARK: - OpenAI Request Types
struct ChatCompletionRequest: Codable, Sendable {
let model: String
let messages: [ChatMessage]
var temperature: Double?
var top_p: Double?
var max_tokens: Int?
var n_predict: Int?
var stream: Bool?
var stop: [String]?
}
struct ChatMessage: Codable, Sendable {
let role: String
let content: String?
}
// MARK: - OpenAI Response Types
struct ChatCompletionResponse: Codable, Sendable {
let id: String
let object: String
let created: Int
let model: String
let choices: [ChatCompletionChoice]
let usage: UsageInfo
}
struct ChatCompletionChoice: Codable, Sendable {
let index: Int
let message: ChatResponseMessage
let finish_reason: String
}
struct ChatResponseMessage: Codable, Sendable {
let role: String
let content: String
}
struct UsageInfo: Codable, Sendable {
let prompt_tokens: Int
let completion_tokens: Int
let total_tokens: Int
}
// MARK: - Streaming Types
struct ChatCompletionChunk: Codable, Sendable {
let id: String
let object: String
let created: Int
let model: String
let choices: [ChunkChoice]
}
struct ChunkChoice: Codable, Sendable {
let index: Int
let delta: DeltaContent
let finish_reason: String?
}
struct DeltaContent: Codable, Sendable {
let role: String?
let content: String?
}
// MARK: - Model List Types
struct ModelsListResponse: Codable, Sendable {
let object: String
let data: [ModelData]
}
struct ModelData: Codable, Sendable {
let id: String
let object: String
let created: Int
let owned_by: String
}
// MARK: - Health / Error Types
struct HealthResponse: Codable, Sendable {
let status: String
}
struct ErrorDetail: Codable, Sendable {
let message: String
let type: String
let code: String?
}
struct ErrorResponse: Codable, Sendable {
let error: ErrorDetail
}

2
web-app/src/components/ai-elements/tool.tsx
View File

@@ -108,7 +108,7 @@ export const ToolHeader = memo(
return (
<CollapsibleTrigger
className={cn(
'flex w-full items-center gap-2 text-muted-foreground text-sm transition-colors capitalize',
'cursor-pointer flex w-full items-center gap-2 text-muted-foreground text-sm transition-colors capitalize', !isOpen && 'hover:bg-secondary',
className
)}
>

1
web-app/src/constants/localStorage.ts
View File

@@ -27,6 +27,7 @@ export const localStorageKey = {
recentSearches: 'recent-searches',
janModelPromptDismissed: 'jan-model-prompt-dismissed',
agentMode: 'agent-mode',
latestJanModel: 'latest-jan-model',
}
export const CACHE_EXPIRY_MS = 1000 * 60 * 60 * 24

1
web-app/src/constants/models.ts
View File

@@ -2,7 +2,6 @@
* Model-related constants
*/
export const NEW_JAN_MODEL_HF_REPO = 'janhq/Jan-v3-4B-base-instruct-GGUF'
export const JAN_CODE_HF_REPO = 'janhq/Jan-Code-4b-Gguf'
export const DEFAULT_MODEL_QUANTIZATIONS = ['iq4_xs', 'q4_k_m']

45
web-app/src/containers/PromptJanModel.tsx
View File

@@ -3,12 +3,9 @@ import { useJanModelPromptDismissed } from '@/hooks/useJanModelPrompt'
import { useServiceHub } from '@/hooks/useServiceHub'
import { useDownloadStore } from '@/hooks/useDownloadStore'
import { useGeneralSetting } from '@/hooks/useGeneralSetting'
import { useEffect, useState, useMemo, useCallback, useRef } from 'react'
import type { CatalogModel } from '@/services/models/types'
import {
NEW_JAN_MODEL_HF_REPO,
SETUP_SCREEN_QUANTIZATIONS,
} from '@/constants/models'
import { useMemo } from 'react'
import { SETUP_SCREEN_QUANTIZATIONS } from '@/constants/models'
import { useLatestJanModel } from '@/hooks/useLatestJanModel'
export function PromptJanModel() {
@@ -18,35 +15,7 @@ export function PromptJanModel() {
useDownloadStore()
const huggingfaceToken = useGeneralSetting((state) => state.huggingfaceToken)
const [janNewModel, setJanNewModel] = useState<CatalogModel | null>(null)
const [isLoading, setIsLoading] = useState(true)
const fetchAttempted = useRef(false)
const fetchJanModel = useCallback(async () => {
if (fetchAttempted.current) return
fetchAttempted.current = true
try {
const repo = await serviceHub
.models()
.fetchHuggingFaceRepo(NEW_JAN_MODEL_HF_REPO, huggingfaceToken)
if (repo) {
const catalogModel = serviceHub
.models()
.convertHfRepoToCatalogModel(repo)
setJanNewModel(catalogModel)
}
} catch (error) {
console.error('Error fetching Jan Model:', error)
} finally {
setIsLoading(false)
}
}, [serviceHub, huggingfaceToken])
useEffect(() => {
fetchJanModel()
}, [fetchJanModel])
const { model: janNewModel, loading: isLoading } = useLatestJanModel()
const defaultVariant = useMemo(() => {
if (!janNewModel) return null
@@ -91,14 +60,14 @@ export function PromptJanModel() {
setDismissed(true)
}
if (isLoading) return null
if (isLoading || !janNewModel) return null
return (
<div className="fixed bottom-4 right-4 z-50 p-4 shadow-lg bg-background w-4/5 md:w-100 border rounded-lg">
<div className="flex items-center gap-2">
<img src="/images/jan-logo.png" alt="Jan" className="size-5" />
<h2 className="font-medium">
Jan v3 Model
{janNewModel?.display_name ?? janNewModel?.model_name ?? 'Jan Model'}
{defaultVariant && (
<span className="text-muted-foreground">
{' '}
@@ -108,7 +77,7 @@ export function PromptJanModel() {
</h2>
</div>
<p className="mt-2 text-sm text-muted-foreground">
Get started with Jan v3, our recommended local AI model optimized for your device.
Get started with {janNewModel?.display_name ?? 'Jan'}, our recommended local AI model optimized for your device.
</p>
<div className="mt-4 flex justify-end space-x-2">
<Button

41
web-app/src/containers/SetupScreen.tsx
View File

@@ -7,11 +7,8 @@ import { useDownloadStore } from '@/hooks/useDownloadStore'
import { useServiceHub } from '@/hooks/useServiceHub'
import { useEffect, useMemo, useCallback, useState, useRef } from 'react'
import { AppEvent, events } from '@janhq/core'
import type { CatalogModel } from '@/services/models/types'
import {
NEW_JAN_MODEL_HF_REPO,
SETUP_SCREEN_QUANTIZATIONS,
} from '@/constants/models'
import { SETUP_SCREEN_QUANTIZATIONS } from '@/constants/models'
import { useLatestJanModel } from '@/hooks/useLatestJanModel'
import { toast } from 'sonner'
import { Button } from '@/components/ui/button'
import { IconEye, IconSquareCheck } from '@tabler/icons-react'
@@ -92,37 +89,19 @@ function SetupScreen() {
const llamaProvider = getProviderByName('llamacpp')
const [quickStartInitiated, setQuickStartInitiated] = useState(false)
const [quickStartQueued, setQuickStartQueued] = useState(false)
const [janNewModel, setJanNewModel] = useState<CatalogModel | null>(null)
const {
model: janNewModel,
error: metadataFetchFailed,
fetchLatestJanModel,
} = useLatestJanModel()
const [supportedVariants, setSupportedVariants] = useState<
Map<string, 'RED' | 'YELLOW' | 'GREEN' | 'GREY'>
>(new Map())
const [metadataFetchFailed, setMetadataFetchFailed] = useState(false)
const supportCheckInProgress = useRef(false)
const checkedModelId = useRef<string | null>(null)
const [isSupportCheckComplete, setIsSupportCheckComplete] = useState(false)
const huggingfaceToken = useGeneralSetting((state) => state.huggingfaceToken)
const fetchJanModel = useCallback(async () => {
setMetadataFetchFailed(false)
try {
const repo = await serviceHub
.models()
.fetchHuggingFaceRepo(NEW_JAN_MODEL_HF_REPO, huggingfaceToken)
if (repo) {
const catalogModel = serviceHub
.models()
.convertHfRepoToCatalogModel(repo)
setJanNewModel(catalogModel)
} else {
setMetadataFetchFailed(true)
}
} catch (error) {
console.error('Error fetching Jan Model V2:', error)
setMetadataFetchFailed(true)
}
}, [serviceHub, huggingfaceToken])
// Check model support for variants when janNewModel is available
useEffect(() => {
const checkModelSupport = async () => {
@@ -188,8 +167,8 @@ function SetupScreen() {
}, [janNewModel, serviceHub])
useEffect(() => {
fetchJanModel()
}, [fetchJanModel])
fetchLatestJanModel(true)
}, [fetchLatestJanModel])
const defaultVariant = useMemo(() => {
if (!janNewModel) return null
@@ -459,7 +438,7 @@ function SetupScreen() {
<div className="flex flex-col w-full h-full justify-center">
<div className="flex flex-1 items-center justify-between">
<h1 className="font-semibold text-sm mb-1">
<span>Jan v3</span>&nbsp;<span className='text-xs text-muted-foreground'>· {defaultVariant?.file_size}</span>
<span>{janNewModel?.display_name ?? janNewModel?.model_name ?? 'Jan Model'}</span>&nbsp;<span className='text-xs text-muted-foreground'>· {defaultVariant?.file_size}</span>
</h1>
{(isDownloading) && (
<div className="flex items-center gap-1.5 text-xs text-muted-foreground">

33
web-app/src/hooks/useJanModelPrompt.ts
View File

@@ -3,6 +3,7 @@ import { create } from 'zustand'
import { createJSONStorage, persist } from 'zustand/middleware'
import { useModelProvider } from './useModelProvider'
import { useDownloadStore } from './useDownloadStore'
import { useLatestJanModel } from './useLatestJanModel'
import { predefinedProviders } from '@/constants/providers'
export type JanModelPromptDismissedState = {
@@ -30,6 +31,7 @@ export const useJanModelPrompt = () => {
const { dismissed, setDismissed } = useJanModelPromptDismissed()
const { getProviderByName, providers } = useModelProvider()
const { localDownloadingModels } = useDownloadStore()
const latestModel = useLatestJanModel((state) => state.model)
const llamaProvider = getProviderByName('llamacpp')
const setupCompleted =
@@ -54,26 +56,31 @@ export const useJanModelPrompt = () => {
})
const isOnSetupScreen = !hasValidProviders
// Check if any Jan v3 variant is downloaded
const isJanModelDownloaded =
llamaProvider?.models.some(
(m: { id: string }) =>
m.id.toLowerCase().includes('jan-v3') ||
m.id.toLowerCase().includes('jan_v3')
) ?? false
// Check if currently downloading
const isDownloading = Array.from(localDownloadingModels).some(
(id) =>
id.toLowerCase().includes('jan-v3') ||
id.toLowerCase().includes('jan_v3')
// Build set of known quant model IDs from the latest Jan model
const latestModelQuantIds = new Set(
latestModel?.quants?.map((q) => q.model_id.toLowerCase()) ?? []
)
// Check if any variant of the latest Jan model is downloaded
const isJanModelDownloaded =
latestModelQuantIds.size > 0 &&
(llamaProvider?.models.some(
(m: { id: string }) => latestModelQuantIds.has(m.id.toLowerCase())
) ?? false)
// Check if currently downloading any variant
const isDownloading =
latestModelQuantIds.size > 0 &&
Array.from(localDownloadingModels).some(
(id) => latestModelQuantIds.has(id.toLowerCase())
)
const showJanModelPrompt =
isTargetVersion &&
!isOnSetupScreen &&
!setupCompleted &&
!dismissed &&
latestModel != null &&
!isJanModelDownloaded &&
!isDownloading

65
web-app/src/hooks/useLatestJanModel.ts Normal file
View File

@@ -0,0 +1,65 @@
import { create } from 'zustand'
import { persist, createJSONStorage } from 'zustand/middleware'
import { localStorageKey, CACHE_EXPIRY_MS } from '@/constants/localStorage'
import { getServiceHub } from '@/hooks/useServiceHub'
import type { CatalogModel } from '@/services/models/types'
type LatestJanModelState = {
model: CatalogModel | null
lastFetchedAt: number | null
loading: boolean
error: boolean
fetchLatestJanModel: (force?: boolean) => Promise<void>
}
export const useLatestJanModel = create<LatestJanModelState>()(
persist(
(set, get) => ({
model: null,
lastFetchedAt: null,
loading: false,
error: false,
fetchLatestJanModel: async (force = false) => {
const { lastFetchedAt, loading } = get()
if (loading) return
if (
!force &&
lastFetchedAt &&
Date.now() - lastFetchedAt < CACHE_EXPIRY_MS
) {
return
}
set({ loading: true, error: false })
try {
const result = await getServiceHub()
.models()
.fetchLatestJanModel()
if (result) {
set({
model: result,
lastFetchedAt: Date.now(),
loading: false,
})
} else {
set({ error: true, loading: false })
}
} catch {
set({ error: true, loading: false })
}
},
}),
{
name: localStorageKey.latestJanModel,
storage: createJSONStorage(() => localStorage),
partialize: (state) => ({
model: state.model,
lastFetchedAt: state.lastFetchedAt,
}),
}
)
)

2
web-app/src/hooks/useLocalApiServer.ts
View File

@@ -51,7 +51,7 @@ export const useLocalApiServer = create<LocalApiServerState>()(
set({ defaultModelLocalApiServer: model }),
lastServerModels: [],
setLastServerModels: (models) => set({ lastServerModels: models }),
serverHost: '0.0.0.0',
serverHost: '127.0.0.1',
setServerHost: (value) => set({ serverHost: value }),
// Use port 0 (auto-assign) for mobile to avoid conflicts, 1337 for desktop
serverPort: (typeof window !== 'undefined' && (window as { IS_ANDROID?: boolean }).IS_ANDROID) || (typeof window !== 'undefined' && (window as { IS_IOS?: boolean }).IS_IOS) ? 0 : 1337,

8
web-app/src/routes/__root.tsx
View File

@@ -15,6 +15,7 @@ import { useAnalytic } from '@/hooks/useAnalytic'
import { PromptAnalytic } from '@/containers/analytics/PromptAnalytic'
import { useJanModelPrompt } from '@/hooks/useJanModelPrompt'
import { PromptJanModel } from '@/containers/PromptJanModel'
import { useLatestJanModel } from '@/hooks/useLatestJanModel'
import { AnalyticProvider } from '@/providers/AnalyticProvider'
import { useLeftPanel } from '@/hooks/useLeftPanel'
import ToolApproval from '@/containers/dialogs/ToolApproval'
@@ -43,6 +44,13 @@ const AppLayout = () => {
width: sidebarWidth,
setLeftPanelWidth,
} = useLeftPanel()
const fetchLatestJanModel = useLatestJanModel(
(state) => state.fetchLatestJanModel
)
useEffect(() => {
fetchLatestJanModel()
}, [fetchLatestJanModel])
return (
<div className="bg-neutral-50 dark:bg-background size-full relative">

20
web-app/src/routes/threads/$threadId.tsx
View File

@@ -145,7 +145,9 @@ function ThreadDetail() {
// context-limit hit, so the user sees it instead of a blank gap.
const [pendingContinueMessage, setPendingContinueMessage] =
useState<UIMessage | null>(null)
const [isAutoIncreasingContext, setIsAutoIncreasingContext] = useState(false)
const [autoIncreaseAttempts, setAutoIncreaseAttempts] = useState(0)
const MAX_AUTO_INCREASE_ATTEMPTS = 3
const isAutoIncreasingContext = autoIncreaseAttempts > 0 && autoIncreaseAttempts < MAX_AUTO_INCREASE_ATTEMPTS
const [contextLimitError, setContextLimitError] = useState<Error | null>(null)
// Refs so onFinish (captured in closure) always calls the latest callbacks
@@ -765,6 +767,9 @@ function ThreadDetail() {
// Increase context length in steps: <8192 -> 8192 -> 32768 -> x1.5
const currentCtxLen =
(model.settings?.ctx_len?.controller_props?.value as number) ?? 8192
const maxCtxLen =
(model.settings?.ctx_len?.controller_props?.max as number) || 131072
let newCtxLen: number
if (currentCtxLen < 8192) {
newCtxLen = 8192
@@ -774,6 +779,12 @@ function ThreadDetail() {
newCtxLen = Math.round(currentCtxLen * 1.5)
}
newCtxLen = Math.min(newCtxLen, maxCtxLen)
if (newCtxLen <= currentCtxLen) {
setContextLimitError(new Error(OUT_OF_CONTEXT_SIZE))
return
}
const updatedModel = {
...model,
settings: {
@@ -818,6 +829,7 @@ function ThreadDetail() {
)
useEffect(() => {
if (!error || agentModeActive) return
if (autoIncreaseAttempts >= MAX_AUTO_INCREASE_ATTEMPTS) return
const autoIncrease =
selectedModel?.settings?.auto_increase_ctx_len?.controller_props?.value ??
true
@@ -829,7 +841,7 @@ function ThreadDetail() {
error.message?.toLowerCase().includes('limit'))) ||
error.message === OUT_OF_CONTEXT_SIZE
if (isContextError) {
setIsAutoIncreasingContext(true)
setAutoIncreaseAttempts((prev) => prev + 1)
handleContextSizeIncrease()
}
}, [error]) // eslint-disable-line react-hooks/exhaustive-deps
@@ -838,8 +850,8 @@ function ThreadDetail() {
if (status === 'streaming' || status === 'submitted') {
setContextLimitError(null)
}
if (isAutoIncreasingContext && (status === 'streaming' || status === 'error')) {
setIsAutoIncreasingContext(false)
if (status === 'streaming' && autoIncreaseAttempts > 0) {
setAutoIncreaseAttempts(0)
}
if (status === 'error' && pendingContinueMessage) {
setPendingContinueMessage(null)

21
web-app/src/services/models/default.ts
View File

@@ -60,6 +60,27 @@ export class DefaultModelsService implements ModelsService {
}
}
async fetchLatestJanModel(): Promise<CatalogModel | null> {
try {
const response = await fetch(LATEST_JAN_MODEL_URL)
if (!response.ok) {
console.error(
`Failed to fetch latest Jan model: ${response.status} ${response.statusText}`
)
return null
}
const data = await response.json()
const model: CatalogModel = Array.isArray(data) ? data[0] : data
return model ?? null
} catch (error) {
console.error('Error fetching latest Jan model:', error)
return null
}
}
async fetchHuggingFaceRepo(
repoId: string,
hfToken?: string

2
web-app/src/services/models/types.ts
View File

@@ -27,6 +27,7 @@ export interface SafetensorsFile {
export interface CatalogModel {
model_name: string
display_name?: string
description: string
library_name?: string
developer?: string
@@ -105,6 +106,7 @@ export interface ModelsService {
getModel(modelId: string): Promise<modelInfo | undefined>
fetchModels(): Promise<modelInfo[]>
fetchModelCatalog(): Promise<ModelCatalog>
fetchLatestJanModel(): Promise<CatalogModel | null>
fetchHuggingFaceRepo(
repoId: string,
hfToken?: string

3
web-app/src/services/providers/tauri.ts
View File

@@ -45,8 +45,11 @@ export class TauriProvidersService extends DefaultProvidersService {
}
}).filter(Boolean)
// TODO: Re-enable foundation-models once migrated to apple-foundation-models crate
const hiddenProviders = new Set(['foundation-models'])
const runtimeProviders: ModelProvider[] = []
for (const [providerName, value] of EngineManager.instance().engines) {
if (hiddenProviders.has(providerName)) continue
const models = await value.list() ?? []
const provider: ModelProvider = {
active: false,

1
web-app/src/types/global.d.ts vendored
View File

@@ -20,6 +20,7 @@ declare global {
declare const POSTHOG_KEY: string
declare const POSTHOG_HOST: string
declare const MODEL_CATALOG_URL: string
declare const LATEST_JAN_MODEL_URL: string
declare const AUTO_UPDATER_DISABLED: boolean
declare const UPDATE_CHECK_INTERVAL_MS: number
declare const GA_MEASUREMENT_ID: string

3
web-app/src/types/modelProviders.d.ts vendored
View File

@@ -8,6 +8,9 @@ type ControllerProps = {
options?: Array<{ value: number | string; name: string }>
input_actions?: string[]
recommended?: string
min?: number
max?: number
step?: number
}
/**

3
web-app/vite.config.ts
View File

@@ -96,6 +96,9 @@ export default defineConfig(({ mode }) => {
MODEL_CATALOG_URL: JSON.stringify(
'https://raw.githubusercontent.com/janhq/model-catalog/main/model_catalog_v2.json'
),
LATEST_JAN_MODEL_URL: JSON.stringify(
'https://raw.githubusercontent.com/janhq/model-catalog/main/latest_jan_model.json'
),
AUTO_UPDATER_DISABLED: JSON.stringify(
env.AUTO_UPDATER_DISABLED === 'true'
),