repository·dev2·Indexed 20 days ago
https://github.com/bigsweetpotatostudio/hyperchatHyperChat is an open-source local AI Agent platform using a 'Configuration as Code' approach. It features a dual-mode architecture consisting of a Web Frontend for multi-workspace collaboration and a CLI Frontend for agent-first interaction. The platform supports the Model Context Protocol (MCP), allowing developers to manage AI capabilities, memory, and tools locally. It utilizes an agent-centric model where each agent is a self-contained application with its own configuration (agent.yaml), memory, and toolsets.
HyperChat is an open-source chat client designed to provide a high-quality chat experience through two core pillars:
Configure how the queue reacts to task failures using errorStrategy:
'stop': Stops the entire queue; all pending tasks are rejected.'continue': Continues processing other tasks (default).'retry': Retries the failed task until retryCount is reached.Set the timeout option (in milliseconds) to automatically reject tasks that run longer than the specified duration.
// Retry strategy
const queue = new TaskQueue({
concurrency: 1,
errorStrategy: 'retry',
retryCount: 3
});
// Timeout control
const timeoutQueue = new TaskQueue({
concurrency: 1,
timeout: 5000 // 5 seconds
});
try {
await timeoutQueue.add(async () => {
await new Promise(resolve => setTimeout(resolve, 10000));
return 'Done';
});
} catch (error) {
console.log('Task timed out');
}HyperChat implements an "Everything is MCP" concept, providing four primary categories of built-in MCP tools to enhance Agent capabilities:
Call Agent and New Task functionality.HyperChat has transitioned from a 'Workspace-centric' model to an 'Agent-centric' model. In this architecture, the Workspace acts primarily as a discovery path for Agents, while each Agent is a fully self-contained AI application.
.hyperchat/agents/{agent_name}/.HyperChat 2.0 uses a dual-mode architecture designed for different interaction needs:
hyperchat serve (accessible at http://localhost:16100).hyperchat <command> (e.g., hyperchat "hello").HyperChat settings use the JSONC format, which allows for comments within the configuration files.
Technical Implementation Details:
Zod for runtime validation to ensure type safety.hyper_knowledge_base is a built-in MCP designed to manage and add content to a knowledge base. Note that this feature is currently in an experimental version and is subject to future improvements.HyperChat 2.0 uses a dual-layer architecture to optimize for different usage scenarios:
Web Layer (Workspace-centric): Designed for multi-workspace collaboration. It uses a WorkspaceManager-enhanced to manage multiple workspace instances, provides a unified MCP (Model Context Protocol) service pool at the workspace level, and uses SSE for real-time data synchronization. It is ideal for project-level management and team collaboration.
CLI Layer (Agent-centric): Designed for rapid, agent-first interaction. It allows direct access to Agent instances, bypassing workspace initialization. It features on-demand toolchain loading, independent agent memory/chat history, and is optimized for scripts and automation without UI overhead.
Configuration follows a 5-level priority merging mechanism (Global + Workspace).
// Web Layer: Workspace-level MCP management
const workspace = workspaceManager.get(workspacePath);
const mcpManager = workspace.getMcpManager();
const client = mcpManager.getClient(clientName);
// CLI Layer: Agent-first MCP access
const agentInstance = workspace.getAgentInstance(agentName);
const client = agentInstance.getMCPClient(clientName);HyperChat uses a two-tier settings architecture to balance global preferences with project-specific configurations:
App Settings (Global)
AppData/app-settings.jsonc.Workspace Settings (Local)
[Workspace]/.hyperchat/settings.jsonc.Settings are applied in the following order of precedence (highest to lowest): Workspace Settings > App Settings > System Defaults.
Workspace settings will override any matching keys defined in App Settings.
HyperChat is built on the philosophy that an Agent = Intelligence (LLM) + Tool + Memory. The software focuses on creating Agents that can interact with tools and perform tasks.
Key capabilities include:
Gemini Think Agent to solve complex reasoning tasks using built-in MCP capabilities.Workspace settings are persisted locally within each workspace directory.
.hyperchat/settings.jsonc inside the workspace folder. The .jsonc format is used to support JSON with comments.settings.schema.json file. If you use a compatible editor, you will receive intelligent autocompletion and validation while editing the settings file directly.Set the concurrency option to a value greater than 1 to execute multiple tasks in parallel.
When adding a task via .add(task, priority), you can specify a priority level. Lower numerical values indicate higher priority. The queue automatically maintains the execution order based on these values.
You can change the number of concurrent tasks at runtime using .setConcurrency(n) and check the current value with .getConcurrency().
// Concurrent execution example
const concurrentQueue = new TaskQueue({ concurrency: 3 });
const tasks = Array.from({ length: 5 }, (_, i) =>
concurrentQueue.add(async () => {
return `Task ${i + 1} complete`;
})
);
const results = await Promise.all(tasks);
// Priority example
const lowPriorityTask = queue.add(async () => 'low', 10);
const highPriorityTask = queue.add(async () => 'high', 1);
// Dynamic adjustment
queue.setConcurrency(4);
console.log(queue.getConcurrency()); // 4