Opensquilla

OpenSquilla is an LLM agent orchestration framework designed to coordinate multi-step AI workflows and tool execution using directed acyclic graphs. It functions as a centralized system for managing specialized skill packages and executing complex reasoning sequences.

The project distinguishes itself through a routing gateway that directs tasks to different AI providers based on complexity, cost, and performance. It utilizes a multi-tier AI memory system that organizes working, episodic, and semantic knowledge using local embeddings and SQLite, alongside a secure execution sandbox that isolates agent-generated code via risk-based permission profiles.

The platform covers a broad range of capabilities, including multi-channel deployment to web and messaging platforms, automated task scheduling via cron, and a Model Context Protocol bridge for connecting to external tools. It also provides comprehensive monitoring and observability tools for tracking token costs, auditing runtime decisions, and managing a catalog of reusable skills.

The system includes command-line utilities for workspace initialization and skill lifecycle management.

Features

Agentic LLM Frameworks - Provides a comprehensive platform for building and managing autonomous LLM agents with support for tool use and multi-step workflows.
LLM Reasoning Workflows - Coordinates complex multi-step AI tasks and tool execution using directed acyclic graphs for reasoning workflows.
Multi-Layer Memory Architectures - Organizes agent knowledge across working, episodic, and semantic layers with automated consolidation for long-term recall.
Sub-Agent Task Delegation - Offloads specific sub-tasks to bounded agents to maintain a clean conversation history in the primary process.
Agent Skill Catalogs - Maintains a curated registry of reusable tools and workflows that agents can autonomously discover and inject.
Agentic Orchestrators - Sequences multi-step tasks and bounded generations to decompose and solve complex reasoning problems.
Skill Packaging - Provides mechanisms to bundle agent instructions, templates, and schemas into reusable, versioned skill packages.
Three-Tier Memory Architectures - Implements a three-tier memory architecture using working, episodic, and semantic layers with automated consolidation.
AI Agent Skills - Provides a framework for defining, installing, and managing reusable instructions and capabilities for autonomous agents.
AI Execution Sandboxes - Deno AI Agent runs code execution within isolated policy tiers and monitors for permission denials.
Temporal Decay Layers - Stores information across working, episodic, semantic, and raw layers with temporal decay and consolidation.
Model Request Routing - Deno AI Agent directs requests to various AI providers or mixed-model profiles based on predefined routing modes to optimize quality.
Cost-Aware Model Routers - Deno AI Agent directs simple requests to cheaper models and reserves high-capability models for complex tasks to reduce spending.
Semantic Complexity Routing - Deno AI Agent analyzes prompt complexity and semantic embeddings on-device to route requests to the cheapest capable provider.
MCP Skill Loaders - Enables on-demand loading of specialized coding and document processing skills via the Model Context Protocol.
Unified Provider Interfaces - Provides a unified interface layer that allows switching between different AI model backends without code changes.
Conversational Session Management - Deno AI Agent persists chat history and state to allow resuming or deleting conversations across interfaces.
Dynamic Skill Injection - Dynamically injects targeted operating instructions and scripts into the agent runtime based on specific task requirements.
LLM Gateways - Centralizes access to multiple LLM providers through a single gateway with primary and fallback routing logic.
LLM Provider Integrations - Provides a centralized configuration surface for connecting and authenticating with multiple LLM providers.
Model Provider Management - Deno AI Agent distributes tasks among different model backends and tiers to balance performance and operational cost.
Real-Time Web Search Integrations - Deno AI Agent fetches real-time information from multiple search providers through a guarded interface to prevent SSRF.
AI Gateways - Hosts a centralized server that routes tasks to various AI providers through foreground and background processes.
Human-in-the-Loop Approvals - Deno AI Agent pauses sensitive actions for manual review based on permission modes to prevent unauthorized changes.
Subagent Orchestrations - Spawns depth-bounded worker agents with isolated contexts to decompose and solve complex tasks.
Reusable Agent Workflows - Packages recurring task sequences into reusable skills that can be executed across different sessions.
Agent Memory Persistence - Saves and retrieves agent state and project memories to non-volatile storage for cross-session continuity.
Hybrid Vector-Keyword Indexing - Implements a hybrid indexing system combining dense vector embeddings with keyword indices for private local knowledge retrieval.
Agent Memory Management - Inspects and manages the long-term memory used by agents to maintain context across multiple sessions.
Multi-Tier Memory Systems - Implements a hierarchical memory architecture to store conversation data across different latency and persistence tiers.
Abstraction Tiering - Organizes content into tiered levels of detail to optimize token usage for LLM consumption.
Agent Model Profiles - Implements named runtime profiles with specific defaults to isolate different agent personas and work streams.
Multi-Step Workflow Orchestration - Organizes recurring tasks into reusable and inspectable sequences that can be autonomously discovered.
Agentic Workflow Orchestrators - Coordinates multi-step agentic tasks through standardized protocols triggered by manual commands or detected user intent.
Code Execution Sandboxes - Provides secure, isolated environments to execute agent-generated scripts and system commands while preventing host system access.
Multi-Channel AI Deployments - Connects a single agent configuration across web interfaces and multiple messaging platforms for unified access.
Agent Skill DAG Execution - Uses directed acyclic graphs to coordinate reasoning and tool execution order for repeatable task sequences.
Execution Sandboxes - Isolates agent-generated code and tool execution within sandboxes using risk-based permission profiles to protect the host system.
Isolated Code Execution - Deno AI Agent runs agent actions in restricted environments with risk-based policies to prevent unauthorized system access.
AI Agent Permissions - Deno AI Agent controls the level of execution authority granted to tasks to restrict elevated operations.
File System Access Controls - Deno AI Agent confines file and shell operations to a specific workspace to prevent accidental host writes.
System Integrity Guards - Deno AI Agent protects system integrity through risk-level metadata, recursion guards, and input sanitization.
Complexity-Based Routers - Deno AI Agent directs queries to different providers based on task complexity to minimize token costs.
Workflow Orchestration - Coordinates complex reasoning steps and tool dependencies using directed acyclic graphs to manage multi-step AI workflows.
Agent Interaction Persistence - Maintains persistent agent definitions and dedicated instructions to ensure continuity across separate workspaces and recurring roles.
Messaging Platform Deployments - Routes a single agent runtime to multiple chat services for consistent interaction across different platforms.
Agent Memory Storage - Provides direct database storage solutions for agent memory using on-device embeddings and markdown notes.
Session State Persistence - Stores transcripts and session state in SQLite to enable workspace management and subagent spawning.
Skill Lifecycle Management - Manages the installation, updating, and removal of reusable AI skill packages via a command-line interface.
Token-Efficient Task Execution - Optimizes token efficiency by loading only the necessary instruction packages and scripts for a specific task.
Agent Third-Party Integrations - Connects core agent capabilities to third-party platforms such as Slack, Discord, and Telegram.
Document Generation Skills - Authors structured documents including spreadsheets, presentations, and PDFs using specialized AI agent skills.
AI Workflow Management - Reviews and accepts draft workflow definitions before promoting them to managed system skills.
User Information Collection - Pauses workflows to gather structured data from users based on defined schemas to satisfy request requirements.
Context Persistence - Implements mechanisms for maintaining historical reasoning traces and conversation state across multi-turn interactions.
Agent Persona Definitions - Defines named agent personas with specific system prompts, default models, and workspaces to standardize behavior.
Model Context Protocol - Connects AI agents to external tools and runtime surfaces using the standardized Model Context Protocol.
Tool Execution Approvals - Deno AI Agent pauses sensitive tool executions for human review and provides a mechanism to reset decisions.
Workspace Access Controls - Deno AI Agent limits file system operations to specific directories to ensure writes remain in a scratch area.
Cross-Interface Agent Synchronization - Synchronizes agent state and behavior across web, terminal, and chat interfaces via a centralized gateway.
Agent Tool Integrations - Integrates the filesystem, shell, Git, and web search tools into agent workflows for automated task execution.
Generative AI Configurations - Manages settings for AI providers, routing logic, search engines, and image generation infrastructure.
Chat-Based Image Generations and Analyses - Creates visual assets from text and analyzes uploaded images using vision models within the chat interface.
Guarded Code Modifications - Implements workflows that apply and verify code modifications in a host repository using safety checks.
Interactive Agent Chat Interfaces - Ships terminal and web-based interfaces for real-time human interaction and messaging with AI agents.
Context Summarizations - Deno AI Agent summarizes long conversation histories to preserve recent state and maintain token efficiency.
Manual Context Compactions - Deno AI Agent triggers manual compaction of long sessions to maintain token efficiency and context window limits.
MCP Servers - Implements a Model Context Protocol server to expose managed tools and runtime surfaces to external clients.
Deterministic Tool Executions - Provides the ability to execute CLI-backed skills and tool calls with predictable, state-changing results.
Structured Output Generators - Produces reports, PDFs, and spreadsheets when output is too large or structured for plain text responses.
Prompt Cache Optimizations - Deno AI Agent organizes stable system prompts and tool definitions at the start of requests to increase cache reuse.
Semantic Search - Deno AI Agent combines vector and keyword search using on-device embeddings to keep data private and reduce latency.
Tiered Model Workload Splitting - Deno AI Agent selects the most cost-effective model tier for a given task to prevent expensive models from handling routine work.
Semantic Vector Search - Deno AI Agent performs hybrid vector and keyword searches using on-device inference to keep data private.
Web Search Integrations - Deno AI Agent connects to external search engines using API keys or keyless paths to retrieve real-time information.
One-Shot Automations - Triggers single automation turns or specific tasks within a workspace to complete discrete jobs.
Workflow Protocol Bridges - Exposes internal session workflows as standardized servers via stdio for triggering by external AI clients.
Durable Agent State - Saves stable preferences and project facts as long-term memory to maintain context across sessions.
Meta-Skill Composition - Combines multiple skills and tool calls into single reusable and auditable workflow protocols.
Recurring Prompt Automations - Automates the periodic execution of LLM prompts to generate summaries and reminders.
Briefing Automation - Automates the curation and delivery of scheduled briefings and reports via a fixed timetable.
Structured File Generation - Produces structured file outputs such as HTML pages and reports as a result of executed tasks.
Local Knowledge Base Indexers - Converts local markdown files and codebases into searchable databases for semantic retrieval.
Memory Search Engines - Implements search engines that find relevant stored content by semantic query and session metadata.
Session State Summarizers - Deno AI Agent converts older conversation entries into a durable brief to preserve goals and key results.
Execution Dependency Management - Defines execution order and parallel processing capabilities through a dependency-based workflow system.
Cron Scheduling - Supports defining recurring background tasks for execution using standard cron expressions.
Workflow Progress Monitoring - Deno AI Agent tracks the execution of multi-step tasks through a visual status ribbon.
System Settings Managers - Provides a centralized interface for managing provider configurations, routing rules, and channel permissions.
Agentic Task Workflow Definition - Provides a framework for defining specialized, sequenceable logic blocks and tools to create formal, auditable AI agent workflows.
MCP Client and Server Implementations - Integrates the Model Context Protocol to load task-specific capabilities on demand as either a client or server.
Conditional Execution Routing - Implements dynamic task routing based on output analysis and provides failure-recovery fallback steps.
LLM Token Cost Tracking - Calculates token usage and financial expenditures on a per-turn and per-session basis to monitor spending.
Non-Interactive Agent Turns - Executes AI agent cycles without requiring human prompts or interaction, using configurable timeouts and bounds.
Recurring Job Scheduling - Runs background jobs on a fixed timetable using a built-in cron parser.
Tool Execution Sandboxes - Deno AI Agent restricts automation tasks to a specific workspace using sandboxing to prevent unauthorized access.
Chat Message Gateways - Provides a routing gateway that synchronizes agent behavior and session state across multiple external messaging platforms and web interfaces.
Conversational Channel Integrations - Integrates the AI gateway with external chat platforms to enable agent control from diverse messaging surfaces.
LLM Provider Failovers - Implements a pluggable provider layer with primary and fallback selection to ensure high availability of AI backends.
Resumable Chat Sessions - Provides durable chat and task history storage to allow resuming of previous work sessions.
Agentic Session Persistence - Persists named agent configurations and task progress in external files to ensure continuity across workstreams.
Prompt Injection Defenses - Deno AI Agent escapes metadata and tool results using XML to close common prompt injection attack vectors.
Tool Execution Permissions - Deno AI Agent requires explicit approval for tool execution to ensure actions match the intended task scope.
Session Task Tracking - Tracks conversation history and progress of goals through a durable session tracking system.
SSRF Protections - Deno AI Agent validates outbound URLs and blocks reserved IP ranges to stop server-side request forgery.
Sandbox Security Configurations - Deno AI Agent defines the security state of the execution environment to determine how isolated processes are handled.
Per-Session Workspace Isolations - Deno AI Agent restricts file and shell operations to specific local directories to ensure execution safety.
AI Cost Monitoring - Provides utilities for tracking token usage and model efficiency to optimize overall operational expenses.
API Expenditure Trackers - Records input and output tokens across various sessions and providers to track total model expenditures.
Routing Decision Monitors - Deno AI Agent displays real-time metadata regarding selected models and estimated savings through a terminal interface.
Model Selection Displays - Deno AI Agent renders a heads-up display showing model selection, confidence levels, and cost savings.
AI Agent Activity Monitors - Deno AI Agent inspects detailed run traces to monitor provider and tool events during a session.
Cost and Token Trackers - Tracks token usage and financial costs per session and model to report on AI operational spend.
Token Cost Optimizations - Deno AI Agent disables reasoning billing for simple queries and auto-tunes prompt depth to lower consumption.

lsdefine/GenericAgent

13,017View on GitHub

GenericAgent is an LLM agent framework and autonomous system controller designed to manage local systems, web browsers, and hardware interfaces through action and observation loops. It functions as a tool orchestrator that routes model calls to local executors, enabling the automation of complex tasks on a host machine. The project is distinguished by its self-evolving AI agent capabilities, which convert successful execution paths into reusable procedural scripts and skill trees to reduce future reasoning overhead. It employs a context optimization engine that utilizes layered memory hierarc

modelscope/ms-agent

4,318View on GitHub

ms-agent is an LLM agent framework and multi-agent orchestration system designed to build autonomous entities that combine large language models with tool calling and structured workflows. It serves as a tool integration platform and workflow engine for executing complex tasks through the coordination of specialized agents. The project distinguishes itself through a multimodal agent workflow engine capable of automating the production of text, images, and video. It features a sandboxed code execution environment for running generated code and quantitative data analysis in isolated containers,

phidatahq/phidata

40,734View on GitHub

Phidata is an LLM agent framework and agentic workflow orchestrator used to build autonomous agents that integrate custom data, tools, and memory. It provides a production environment for serving these agents as services via APIs, utilizing server-sent events and websockets for real-time communication. The system distinguishes itself through a human-in-the-loop control layer that requires manual approval and administrative sign-off for specific tool executions. It also implements a multi-tenant AI infrastructure that uses token-based roles to ensure data isolation between different tenants.

Hmbown/DeepSeek-TUI

38,538View on GitHub

DeepSeek-TUI is an AI coding agent orchestrator and framework designed to automate complex programming tasks. It functions as a harness for coordinating AI models that can read source code, edit files, and execute shell commands through automated agent workflows. The system is distinguished by its multi-agent coordination capabilities, which allow for the spawning of parallel sub-agents to handle concurrent investigations or implementation slices. It employs autonomous goal-seeking loops to pursue objectives across multiple turns and utilizes a tool integration gateway to connect models to ex

opensquillaopensquilla

Features