Aden Agent Framework aims to help developers build outcome-oriented, self-adaptive agents. Please find our roadmap here
flowchart TB
%% Main Entity
User([User])
%% =========================================
%% EXTERNAL EVENT SOURCES
%% =========================================
subgraph ExtEventSource [External Event Source]
E_Sch["Schedulers"]
E_WH["Webhook"]
E_SSE["SSE"]
end
%% =========================================
%% SYSTEM NODES
%% =========================================
subgraph WorkerBees [Worker Bees]
WB_C["Conversation"]
WB_SP["System prompt"]
subgraph Graph [Graph]
direction TB
N1["Node"] --> N2["Node"] --> N3["Node"]
N1 -.-> AN["Active Node"]
N2 -.-> AN
N3 -.-> AN
%% Nested Event Loop Node
subgraph EventLoopNode [Event Loop Node]
ELN_L["listener"]
ELN_SP["System Prompt<br/>(Task)"]
ELN_EL["Event loop"]
ELN_C["Conversation"]
end
end
end
subgraph JudgeNode [Judge]
J_C["Criteria"]
J_P["Principles"]
J_EL["Event loop"] <--> J_S["Scheduler"]
end
subgraph QueenBee [Queen Bee]
QB_SP["System prompt"]
QB_EL["Event loop"]
QB_C["Conversation"]
end
subgraph Infra [Infra]
SA["Sub Agent"]
TR["Tool Registry"]
WTM["Write through Conversation Memory<br/>(Logs/RAM/Harddrive)"]
SM["Shared Memory<br/>(State/Harddrive)"]
EB["Event Bus<br/>(RAM)"]
CS["Credential Store<br/>(Harddrive/Cloud)"]
end
subgraph PC [PC]
B["Browser"]
CB["Codebase<br/>v 0.0.x ... v n.n.n"]
end
%% =========================================
%% CONNECTIONS & DATA FLOW
%% =========================================
%% External Event Routing
E_Sch --> ELN_L
E_WH --> ELN_L
E_SSE --> ELN_L
ELN_L -->|"triggers"| ELN_EL
%% User Interactions
User -->|"Talk"| WB_C
User -->|"Talk"| QB_C
User -->|"Read/Write Access"| CS
%% Inter-System Logic
ELN_C <-->|"Mirror"| WB_C
WB_C -->|"Focus"| AN
WorkerBees -->|"Inquire"| JudgeNode
JudgeNode -->|"Approve"| WorkerBees
%% Judge Alignments
J_C <-.->|"aligns"| WB_SP
J_P <-.->|"aligns"| QB_SP
%% Escalate path
J_EL -->|"Report (Escalate)"| QB_EL
%% Pub/Sub Logic
AN -->|"publish"| EB
EB -->|"subscribe"| QB_C
%% Infra and Process Spawning
ELN_EL -->|"Spawn"| SA
SA -->|"Inform"| ELN_EL
SA -->|"Starts"| B
B -->|"Report"| ELN_EL
TR -->|"Assigned"| EventLoopNode
CB -->|"Modify Worker Bee"| WorkerBees
%% =========================================
%% SHARED MEMORY & LOGS ACCESS
%% =========================================
%% Worker Bees Access
Graph <-->|"Read/Write"| WTM
Graph <-->|"Read/Write"| SM
%% Queen Bee Access
QB_C <-->|"Read/Write"| WTM
QB_EL <-->|"Read/Write"| SM
%% Credentials Access
CS -->|"Read Access"| QB_C
Implement the core execution engine where every Agent operates as an isolated, asynchronous graph of nodes.
- Core Node Implementation
- NodeProtocol with JSON parsing utilities (graph/node.py)
- EventLoopNode with LLM conversation management (graph/event_loop_node.py)
- Flexible input/output keys with nullable output handling
- Node wrapper SDK for agent creation
- Tool access layer with MCP integration
- Graph Executor
- Graph traversal execution (graph/executor.py)
- Node transition management
- Error handling and output mapping
- ExecutionResult with success/error status
- Shared Memory Access
- SharedState manager (runtime/shared_state.py)
- Session-based storage (storage/session_store.py)
- Isolation levels: ISOLATED, SHARED, SYNCHRONIZED
- Default Monitoring Hooks
- Performance metrics collection
- Resource usage tracking
- Health check endpoints
Build the standard communication protocol for inter-node messaging and data passing.
- Edge Specifications
- ALWAYS: Always traverse (graph/edge.py)
- ON_SUCCESS: Success-based routing
- ON_FAILURE: Failure-based routing
- CONDITIONAL: Expression-based routing with safe_eval
- LLM_DECIDE: Goal-aware LLM-powered routing
- Event Bus System
- Full event bus implementation (runtime/event_bus.py)
- LLM text deltas, tool calls, node transitions
- Graph-scoped event routing for multi-agent scenarios
- Conversation Management
- NodeConversation tracks message history (graph/conversation.py)
- Tool results, streaming content, metadata support
A separate LLM-powered judge to determine if the workers finish their job.
- Conversation Judge (Level 2)
- Evaluates node completion against success criteria (graph/conversation_judge.py)
- Reads recent conversation and assesses quality
- Returns verdict: ACCEPT or RETRY with confidence scores
- Test Evaluation Judge
- Provider-agnostic (OpenAI, Anthropic, Google Gemini) (testing/llm_judge.py)
- JSON response parsing for structured evaluation
- Multi-Level Judgment Integration
- Judge node integration with event loop
- Automatic retry logic based on judge verdict
- Judge performance monitoring
Develop the distinct behavioral logic for the Queen Bee (Orchestrator), Judge Bee (Evaluator), and Worker Bee (Executor).
- Judge Bee (Evaluator)
- Evaluation criteria framework (graph/goal.py)
- Success/failure determination
- Quality assessment with confidence scores
- Hive Coder Agent (Builder)
- Coder node: forever-alive event loop (agents/hive_coder/nodes/)
- Guardian node: event-driven watchdog for supervised agents
- Tool discovery (discover_mcp_tools)
- Agent aware (list_agents, inspect sessions)
- Post-build testing (run_agent_tests)
- Debugging capabilities (inspect checkpoints, memory)
- Queen Bee (Orchestrator)
- Multi-agent coordination layer
- Task distribution logic
- Dynamic worker agent creation
- Swarm-level goal management
- Worker Bee (Executor)
- Worker taxonomy definition
- Worker agent templates
- Task execution patterns
Implement the Goal Creation Session via the Queen Bee and the dynamic Worker Agent Creation flow.
- Goal Creation Session
- Goal object schema definition (graph/goal.py)
- SuccessCriterion: Measurable success (5+ criteria per goal)
- Constraint: Hard/soft boundaries (time, cost, safety, scope, quality)
- GoalStatus: DRAFT → READY → ACTIVE → COMPLETED/FAILED
- Instruction back and forth in Hive Coder
- Test case generation
- Test case validation for worker agent
- Agent Creation Flow
- Hive Coder reads templates and discovers tools (builder/package_generator.py)
- Generates agent.py, nodes/init.py, config.py
- MCP server configuration discovery
- Dynamic tool binding
- Worker Agent Dynamic Creation
- Template agent initialization from Queen Bee
- Runtime worker instantiation
- Worker lifecycle management
Build robust, local Credential Management interfaces for secure API key handling.
- Unified Credential Store
- Multi-backend storage (credentials/store.py)
- EncryptedFileStorage: Encrypted local storage (~/.hive/credentials)
- EnvVarStorage: Environment variable mapping
- InMemoryStorage: Testing
- HashiCorp Vault: Enterprise secrets (credentials/storage.py)
- Template resolution:
{{cred.key}}patterns - Caching with TTL (default 5 min, configurable)
- Thread-safe operations with RLock
- OAuth2 Providers
- Base provider pattern (credentials/oauth2/)
- HubSpot provider integration
- Lifecycle management (refresh tokens)
- Browser opening for auth flows (tools/credentials/browser.py)
- Aden Sync Provider
- Syncs OAuth2 tokens from Aden authentication server (credentials/aden/)
- Falls back to local storage if Aden unavailable
- Auto-refresh on sync
- Enterprise Secret Managers
- AWS Secrets Manager integration
- Azure Key Vault integration
- Audit logging for compliance/tracking
- Per-environment configuration support
Develop the Sub-agent execution environment for parallel tasks execution. The subagents are designed with isolation for repeatability.
- Multi-Graph Sessions
- Load multiple agent graphs in single session (runtime/agent_runtime.py)
- Shared state between graphs
- Independent execution streams
- Graph lifecycle management (load/unload/start/restart)
- Concurrent Execution Management
- Max concurrent executions configuration
- Isolation levels: isolated, shared, synchronized
- Sub-agent Execution Environment
- Isolated sub-agent runtime environment
- Task isolation mechanisms
- Result aggregation
- Error handling for parallel tasks
- Repeatability guarantees
Implement native browser-integrated automation so agents can take over a browser for auth and agents perform the automation jobs. This node comes with a specific set of tools and system prompts.
- Web Scraping with Playwright
- Headless Chromium launch (tools/web_scrape_tool/)
- Stealth mode via playwright_stealth
- JavaScript rendering with wait-for-domcontentloaded
- CSS selector support
- User-agent spoofing
- Sandbox/automation detection evasion
- Browser Launch Utilities
- Platform-specific browser opening (macOS/Linux/Windows) (tools/credentials/browser.py)
- OAuth2 flow integration
- Full Browser Use Node
- Multi-page automation workflows
- Form filling with vision-guided interactions
- Interactive screenshot capabilities
- Session management across navigations
- Browser-specific tool set
- System prompts for browser tasks
Ship essential framework utilities: Node validation, HITL (Human-in-the-loop pause/approve), and node lifecycle management.
- Node Validation
- Pydantic-based validation
- Schema enforcement
- Output key validation (Level 0)
- Human-in-the-Loop (HITL)
- HITLRequest and HITLResponse protocol (graph/hitl.py)
- Question types: FREE_TEXT, STRUCTURED, SELECTION, APPROVAL, MULTI_FIELD
- Haiku-powered response parsing
- User-friendly display formatting
- Pause/approve workflow
- State saved to checkpoint
- Resume with HITLResponse merged into context
- TUI Integration
- Chat REPL with streaming support (tui/app.py)
- Multi-graph session management
- User presence detection
- Real-time log viewing
- Node Lifecycle Management
- Start/stop/pause/resume in execution stream
- State persistence via checkpoint store
- Recovery mechanisms with checkpoint restore
- Advanced HITL Features
- Callback handlers for custom intervention logic
- Streaming interface for real-time monitoring
- Approval workflows at scale
Port popular tools, and build out the Runtime Log, Audit Trail, Excel, and Email integrations.
- File Operations (36+ tools)
- read_file, write_file, edit_file (builder/package_generator.py)
- list_directory, search_files
- apply_diff / apply_patch for code modification (tools/file_system_toolkits/)
- data_tools (CSV/Excel parsing)
- Web Tools
- Web Search (tools/web_search_tool/)
- Web Scraper (tools/web_scrape_tool/)
- Exa Search (tools/exa_search_tool/)
- News Tool (tools/news_tool/)
- SerpAPI (tools/serpapi_tool/)
- Data Tools
- CSV tools (tools/csv_tool/)
- Excel tools (tools/excel_tool/)
- PDF tools (tools/pdf_read_tool/)
- Vision tool for image analysis (tools/vision_tool/)
- Time tool (tools/time_tool/)
- Communication Tools (8 tools)
- Email tool (tools/email_tool/)
- Gmail tool (tools/gmail_tool/)
- Slack tool (tools/slack_tool/)
- Discord tool (tools/discord_tool/)
- Telegram tool (tools/telegram_tool/)
- Google Docs (tools/google_docs_tool/)
- Google Maps (tools/google_maps_tool/)
- Cal.com (tools/calcom_tool/)
- CRM/API Integrations (5+ tools)
- HubSpot (tools/hubspot_tool/)
- GitHub (tools/github_tool/)
- Apollo (tools/apollo_tool/)
- BigQuery (tools/bigquery_tool/)
- Razorpay (tools/razorpay_tool/)
- Calendar (tools/calendar_tool/)
- Security/Scanning Tools (5 tools)
- DNS Security Scanner (tools/dns_security_scanner/)
- SSL/TLS Scanner (tools/ssl_tls_scanner/)
- Port Scanner (tools/port_scanner/)
- Subdomain Enumerator (tools/subdomain_enumerator/)
- Tech Stack Detector (tools/tech_stack_detector/)
- Runtime & Logging
- Runtime Log Tool (tools/runtime_logs_tool/)
- Runtime Logger with L1/L2/L3 levels (runtime/runtime_logger.py)
- Audit Trail System
- Decision tracing beyond logs
- Compliance reporting
- Historical query capabilities
Simple pure file-based memory management
- Short-Term Memory (STM)
- SharedState manager for in-memory state (runtime/shared_state.py)
- Session-based storage (storage/session_store.py)
- State-based short-term memory layer
- Conversation Memory
- NodeConversation tracks message history (graph/conversation.py)
- Tool results, streaming content, metadata
- Context building for LLM prompts
- Long-Term Memory (LTM)
- Semantic indexing for memory retrieval
- RLM (Retrieval-augmented Long-term Memory) implementation
- Memory persistence beyond session
- Content-based memory search
Integrate a lightweight, persistent DB for long-term memory using the filesystem-as-scratchpad pattern.
- Filesystem as Scratchpad
- File-based persistence layer (storage/)
- Session store implementation
- Data durability guarantees
- Checkpoint System
- Save/restore execution state (storage/checkpoint_store.py)
- TTL-based cleanup
- Async checkpoint support
- Max age configuration
- Message Model & Session Management
- Message class with structured content types
- Session classes for conversation state
- Per-message file persistence
- Migration from monolithic run storage
Enforce session-local memory isolation to prevent data bleed between concurrent agent runs.
- Session Isolation
- Session-local memory implementation (storage/session_store.py)
- Data bleed prevention
- Concurrent run safety
- Isolation levels: ISOLATED, SHARED, SYNCHRONIZED
- State Management
- SharedState with thread-safe operations (runtime/shared_state.py)
- Session-scoped state access
- Context Management
- Message.stream(sessionID) implementation
- Full context building optimization
- Message to model conversion improvements
Implement File I/O support, streaming mode, and allow users to supply custom functions as libraries/nodes.
- File I/O
- File read/write operations (builder/package_generator.py)
- File system navigation
- Directory listing and search
- Execution Streaming
- Real-time event streaming (runtime/execution_stream.py)
- Token-by-token output via event bus
- Tool call streaming
- Custom Tool Integration
- MCP server discovery (builder/package_generator.py)
- Dynamic tool binding
- Custom tool registration
- Streaming Mode Enhancements
- Progressive result delivery optimization
- Backpressure handling
- Custom Function Libraries
- User-supplied function libraries as nodes
- Library versioning and management
- Proactive Memory Compaction
- Overflow detection
- Backward-scanning pruning strategy
- Token tracking integration for compaction decisions
Add semantic search capabilities and an interactive file system for frontend product integration.
- File Search
- search_files tool (builder/package_generator.py)
- Directory traversal
- Semantic Search
- Semantic indexing of files
- Natural language file search
- Content-based retrieval with embeddings
- Interactive File System
- Frontend file browser integration
- Real-time file system updates
- Visual file navigation in GUI
Build the failure recording mechanism and an SDK for defining custom failure conditions.
- Multi-Level Evaluation
- Level 0: Output key validation (all required keys set)
- Level 1: Literal checks (output_contains, output_equals)
- Level 2: Conversation-aware judgment (graph/conversation_judge.py)
- Goal-Based Constraints
- Hard constraints (violation = failure) (graph/goal.py)
- Soft constraints (prefer not to violate)
- Categories: time, cost, safety, scope, quality
- Constraint checking infrastructure
- Success Criteria Definition
- Weighted criteria (0.0-1.0)
- Metrics: output_contains, output_equals, llm_judge, custom
- 90% threshold for goal success
- Test Framework
- TestCase, TestResult, TestStorage classes (testing/)
- LLM-based judgment for semantic evaluation (testing/llm_judge.py)
- Approval CLI for manual approval workflows
- Categorization and test result reporting
- Failure Recording
- Failure capture mechanism
- Failure analysis tools
- Historical failure tracking
- Continuous improvement loop
- Custom Failure Conditions SDK
- SDK for defining custom failure conditions
- Custom evaluator framework extension
- Condition validation DSL
Implement deterministic condition guardrails directly in the node, complete with mitigation tracking and audit logs.
- Goal Constraints (Basic Guardrails)
- Hard/soft constraint definitions (graph/goal.py)
- Constraint checking in goals
- Deterministic Guardrails SDK
- In-node guardrail implementation
- Condition-based guardrails
- Guardrail SDK for custom rules
- Monitoring & Tracking
- Mitigation tracking for violations
- Audit log system for guardrails
- Compliance reporting
- Basic Monitoring Hooks
- Agent node SDK monitoring hooks
- Event hook system for guardrails
- Default monitoring hooks in nodes
Release CLI tools specifically for rapid memory management and credential store editing.
- Main CLI
- Run, info, validate, list commands (cli.py)
- Dispatch mode for batch execution
- Shell mode for interactive use
- Model selection configuration
- Testing CLI
- test-run, test-debug, test-list, test-stats (testing/cli.py)
- Pytest integration
- Test categorization
- TUI (Terminal UI)
- Interactive chat with streaming (tui/app.py)
- Multi-graph management UI
- Log pane for real-time output
- Keyboard shortcuts (Ctrl+C, Ctrl+D, etc.)
- Memory Management CLI
- Memory inspection commands
- Memory cleanup utilities
- Session management commands
- Credential Store CLI
- Interactive credential editing
- Secure credential viewer
- Credential validation tools
- Debugging Tools
- Interactive debugging mode beyond TUI
- Breakpoint support in execution
- Step-through execution
Support user-driven log analysis, basic monitoring hooks from the SDK, and an interactive debugging mode.
- Runtime Logging
- L1 (summary), L2 (detailed), L3 (tool) logging levels (runtime/runtime_logger.py)
- Session logs directory storage
- Audit trail for decision tracing in logs
- Event Bus Monitoring
- Real-time event streaming (runtime/event_bus.py)
- LLM text deltas, tool calls, node transitions
- Graph-scoped event routing
- Log Analysis Tools
- User-driven log analysis (OSS approach)
- Log aggregation utilities
- Log visualization tools
- Monitoring Hooks
- Basic observability hooks from SDK
- Performance metrics collection
- Health checks system
- Token Tracking
- Reasoning token tracking
- Cache token tracking
- Token metrics in compaction logic
Write the Quick Start guide, detailed tool usage documentation, and set up the MVP README examples.
- Documentation
- Quick start guide
- Goal creation guide
- Agent creation guide
- README with examples
- Contributing guidelines
- GitHub Page setup
- Tool Usage Documentation
- Comprehensive tool documentation
- Tool integration examples
- Best practices guide
- Video Content
- Introduction video
- Tutorial videos
- Example Agents
- Knowledge agent template
- Blog writer agent template
- SDR agent template
Standardize the Docker container builds and establish headless backend execution APIs.
- Docker Support
- Python 3.11-slim base image (tools/Dockerfile)
- Playwright Chromium installation
- Non-root user for security
- Health check endpoint
- Volume mount for workspace persistence
- Exposes port 4001 for MCP server
- Agent Runtime
- AgentRuntime: Top-level orchestrator (runtime/agent_runtime.py)
- Multiple entry points (manual, webhook, timer, event, api)
- Concurrent execution management
- State persistence via session store
- Outcome aggregation
- Async Entry Points
- AsyncEntryPointSpec: Webhook, timer, event triggers (graph/edge.py)
- Timer config: cron expressions or interval_minutes
- Event triggers for custom events
- Isolation levels: isolated, shared, synchronized
- Headless Backend Enhancements
- Standardized backend execution APIs
- Frontend attachment interface
- Self-hosted setup guide with examples
Expose basic REST/WebSocket endpoints for external control (Start, Stop, Pause, Resume).
- Webhook Server
- FastAPI-based webhook server (runtime/webhook_server.py)
- Route configuration per entry point
- Optional secret validation
- Graph Lifecycle Management
- Load/unload/start/restart in AgentRuntime
- State persistence
- Recovery mechanisms
- REST API Endpoints
- Start endpoint for agent execution
- Stop endpoint for graceful shutdown
- Pause endpoint for execution suspension
- Resume endpoint for continuation
- Status query endpoint for monitoring
- WebSocket API
- Real-time event streaming to clients
- Bidirectional communication
- Connection management with reconnection
Implement automated test execution, agent version control, and mandatory test-passing for deployment.
- Test Execution
- Test framework with pytest integration (testing/)
- Test result reporting
- Test CLI commands (test-run, test-debug, etc.)
- Automated Testing Pipeline
- CI integration (GitHub Actions, etc.)
- Mandatory test-passing gates
- Coverage reporting
- Version Control
- Agent versioning system
- Semantic versioning for agents
- Version compatibility checks
- Deployment Automation
- Continuous deployment pipeline
- Rollback mechanisms
- Blue-green deployment support
Launch the official PyPI package, Docker Hub image, and the community Discord channel.
- Package Distribution
- Official PyPI package
- Docker Hub image publication
- Version release automation
- Installation documentation
- Community Channels
- Discord channel setup
- Community support structure
- Contribution guidelines enforcement
- Cloud Deployment
- AWS Lambda integration
- GCP Cloud Functions support
- Azure Functions support
- 3rd-party platform integrations
- Self-deploy with orchestrator connection
Ship ~20 ready-to-use templates including GTM Sales, Marketing, Analytics, Training, and Smart Entry agents.
- Hive Coder Agent
- Agent builder template (agents/hive_coder/)
- Guardian node for supervision
- Sales & Marketing Agents
- GTM Sales Agent (workflow automation)
- GTM Marketing Agent (campaign management)
- Lead generation agent
- Email campaign agent
- Social media agent
- Analytics & Insights Agents
- Analytics Agent (data analysis)
- Data processing agent
- Report generation agent
- Dashboard agent
- Training & Education Agents
- Training Agent (onboarding)
- Content creation agent
- Knowledge base agent
- Documentation agent
- Automation & Forms Agents
- Smart Entry / Form Agent (self-evolution emphasis)
- Data validation agent
- Workflow automation agent
- Integration agent
- Additional Templates
- Customer support agent
- Document processing agent
- Scheduling agent
- Research agent
- Code review agent
Build a lightweight local server (e.g., FastAPI or Node) that securely exposes the Hive framework's core Event Bus and Memory Layer to the local browser environment.
- MCP Server Foundation
- FastMCP server implementation (builder/package_generator.py)
- Agent builder tools exposed
- Port 4001 exposed in Docker
- Event Bus Architecture
- Event Bus implementation (runtime/event_bus.py)
- Real-time event streaming
- Graph-scoped event routing
- Local API Gateway
- Lightweight local server (FastAPI or Node)
- Secure authentication layer for browser
- CORS and security configuration
- Event Bus API endpoints for browser access
- Event subscription management for frontend
- Memory Layer API
- Memory read/write endpoints
- Session management API for frontend
- Memory visualization data endpoints
Implement an interactive, drag-and-drop canvas (using libraries like React Flow) to visualize the Worker Graph, Queen Bee, and active execution paths in real-time.
- Graph Visualization
- React Flow integration
- Worker Graph rendering from agent definitions
- Node type visualization (EventLoop, Function, etc.)
- Edge visualization with condition types
- Active execution path highlighting
- Interactive Features
- Drag-and-drop canvas for graph editing
- Node editing capabilities
- Real-time graph updates during execution
- Zoom and pan controls
- Node inspection on click
- Integration with Runtime
- Live execution visualization
- Node state indicators
- Edge traversal animation
Port the existing Terminal User Interface (TUI) into a rich web application, allowing users to interact directly with the Queen Bee / Coding Agent via a browser chat interface.
- TUI Foundation
- Terminal chat interface (tui/app.py)
- Streaming support
- Multi-graph management
- Log pane display
- Keyboard shortcuts
- Web Application
- Modern web UI framework setup (React/Vue/Svelte)
- Responsive design implementation
- Cross-browser compatibility
- Chat Interface
- Browser-based chat UI
- Hive Coder interaction (Queen Bee proxy)
- Coding Agent interface
- Message history and search
- Rich message formatting (markdown, code blocks)
- TUI Feature Parity
- All TUI commands in GUI
- Keyboard shortcuts in browser
- Command palette (Cmd+K style)
Create a UI component to inspect the Shared Memory and Write-Through Conversation Memory, allowing developers to click on any node and see exactly what it is thinking.
- Runtime Logs Tool
- Inspect agent session logs (tools/runtime_logs_tool/)
- Session state retrieval (builder/package_generator.py)
- Memory Inspector UI
- Shared Memory visualization
- Conversation memory view (NodeConversation display)
- Memory search and filter
- Memory timeline view
- Node State Inspection
- Click-to-inspect functionality
- Node thought process display (LLM reasoning)
- State history timeline per node
- Input/output inspection
- Debug Tools
- Memory diff viewer (state changes between nodes)
- State snapshot comparison
- Memory leak detection
Build a dashboard for localized Credential Management (editing the ~/.hive/credentials store safely) and swarm lifecycle management (Start, Pause, Kill, and HITL approvals).
- Credential Management Backend
- CredentialStore with file/env/vault backends (credentials/store.py)
- OAuth2 provider support (credentials/oauth2/)
- Template resolution and caching
- Credential Management Dashboard
- Safe credential editing interface (web UI)
- ~/.hive/credentials store management UI
- Credential validation and testing UI
- Encryption status display
- OAuth2 flow initiation from browser
- Swarm Lifecycle Management
- Start/Stop controls for agents
- Pause/Resume functionality
- Kill process management
- HITL approval interface in browser
- Multi-agent orchestration view
- Monitoring Dashboard
- Active agents display
- Resource usage monitoring (CPU, memory, tokens)
- Performance metrics visualization
- Execution history
Build native frontend configurations to easily connect Open Hive's backend to local open-source inference engines like Ollama, keeping the entire stack offline and private.
- LLM Integration Layer
- Provider-agnostic LLM support via LiteLLM (graph/event_loop_node.py)
- Model configuration in agent definitions
- Local Model Support
- Ollama integration and configuration
- Local LLM configuration UI
- Model selection and management dashboard
- Model performance monitoring
- Offline Mode
- Full offline functionality (no cloud API calls)
- Local-only execution mode flag
- Privacy-first architecture enforcement
- Local model fallback mechanisms
- Model Configuration
- Easy model switching in UI
- Model parameter tuning (temperature, top_p, etc.)
- Performance optimization settings
- Multi-model support (different models per node)
- Model cost tracking for local models
- JavaScript/TypeScript SDK
- TypeScript SDK development
- npm package distribution
- Node.js runtime support
- Browser runtime support
- Platform Compatibility
- Windows support improvements
- macOS optimization
- Linux distribution support
- IDE Integrations
- Claude Code integration
- Cursor integration
- Opencode integration
- Antigravity integration
- Codex CLI integration (in progress)