Skip to content

CHANGELOG.md

Latest commit

 

History

History
410 lines (317 loc) · 30.6 KB

CHANGELOG.md

File metadata and controls

410 lines (317 loc) · 30.6 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.9] - 2026-03-16

Added

  • Mini Pomodoro Timer — Compact countdown widget in the global header bar, visible on all pages when a Pomodoro session is active. Shows progress ring, session type icon, and countdown. Click to navigate to Pomodoro page. Auto-hides when no session is running or when on the Pomodoro page itself.
  • MiniPomodoro component (packages/ui/src/components/MiniPomodoro.tsx) — Self-contained component with WebSocket updates and independent countdown

Fixed

  • Pomodoro Timer Broken on Non-UTC Machines — Root cause: PostgreSQL TIMESTAMP (without timezone) columns + pg driver's default parser interprets stored values as local time. On non-UTC machines, startedAt was shifted backward, causing fetchState() to calculate remaining = 0 and immediately auto-complete every session. Fix: pg.types.setTypeParser(1114) forces UTC interpretation for all timestamp columns.
  • Pomodoro Timer Effect — Removed timeLeft from the timer useEffect dependency array, which was needlessly recreating the interval every second
  • CI: notification-router.test.ts — Test was hitting real PostgreSQL (port 25432) instead of mocks after preferences were migrated to DB-backed storage. Added settingsRepo mock with in-memory Map.
  • Code Formatting — Fixed 88 files with Prettier formatting issues that were failing format:check in CI

0.1.10 - 2026-03-14

Added

  • 6 Workflow Node UI Components — DataStore (cyan), SchemaValidator (orange), Filter (emerald), Map (sky), Aggregate (amber), WebhookResponse (rose) with color-coded canvas nodes, status indicators, and execution timing
  • 6 Node Config Panels — Full right-panel editors for each new node type with field editors, template validation, output alias, retry/timeout
  • LLM Conversation Context Editor — Add/edit/remove multi-turn messages with role selector before main user message
  • LLM Response Format UI — Text/JSON selector with JSON badge on canvas node
  • Execution Progress Bar — Real-time toolbar showing running node name, completed/total count, retry counter
  • Execution Timeline Labels — Node labels instead of raw IDs in both live and historical log views
  • 9 Workflow Integration Tests — Template resolution, condition branching, ForEach body, error propagation, filter/map/aggregate, DataStore persistence

Fixed

  • Workflow Execution — node_complete event now includes retryAttempts; cancelled workflows logged as 'cancelled' instead of 'failed'
  • Circular Dependencies (core) — Broke fragile tool-validation ↔ tools cycle with lazy import (was last remaining dangerous cycle)
  • Circular Dependencies (gateway) — Broke 45+ cycles across 4 root causes: agents↔ws/server (lazy import), tool-providers↔routes (cached lazy executors), webchat-handler↔server (lazy import), normalizers barrel (extracted types.ts)

Changed

  • ToolPalette sidebar now shows all 23 node types
  • NodeSearchPalette includes 6 new node types
  • Node config router dispatches to 23 panel types
  • All normalizers import from types.ts instead of barrel index.ts

Removed

  • Deleted orphan services/_gen.js (console.log stub, unused)

0.1.9 - 2026-03-14

Added

  • 6 New Workflow Nodes — DataStore (key-value persistence), SchemaValidator (JSON schema validation), Filter (array filtering), Map (array transformation), Aggregate (sum/count/avg/min/max/groupBy/flatten/unique), WebhookResponse (HTTP response for webhook triggers)
  • LLM Node Improvements — responseFormat: 'json' for auto-parsed JSON output, conversationMessages for multi-turn context
  • 5 Workflow Templates — GitHub Issue Triage, Data Pipeline, Scheduled Report, Multi-Source Merge, Approval Workflow
  • Webhook Trigger Integration — POST /webhooks/workflow/:path endpoint with HMAC-SHA256 signature validation
  • Approval Recovery — resumeFromApproval() auto-resumes paused workflows when approval is decided
  • Fleet Command Tests — 68 comprehensive tests covering lifecycle, scheduling, task execution, budgets, concurrency
  • Fleet Event-Driven Scheduling — Fleets can now trigger cycles on EventBus events
  • Fleet Shared Context Feedback — Worker outputs automatically feed back into shared context for downstream workers
  • Fleet Session Cleanup — Old completed/failed sessions automatically cleaned on boot

Fixed

  • Cost Tracking (all systems) — calculateExecutionCost() shared utility now populates costUsd in BackgroundAgentRunner, SubagentRunner, FleetWorker, and SoulHeartbeatService; budget enforcement is now functional
  • Fleet Dependency Cascade — Failed tasks now propagate failure to all dependent tasks (no more deadlocked task chains)
  • Fleet Shared Context Mutation — structuredClone() prevents cross-worker context corruption
  • Fleet Cron Scheduling — Now uses getNextRunTime() from core instead of stub 60s fallback
  • Fleet Orphaned Tasks — Tasks stuck in 'running' from crashes are re-queued on fleet start
  • Fleet executed_at Timestamp — Worker execution time now properly saved to DB
  • Workflow DataStore Memory Leak — Added 10K entry limit with LRU eviction and clearDataStore() export
  • Workflow SubWorkflow Auth — Added userId ownership check (prevents cross-user access)
  • Workflow SubWorkflow Abort — Parent abort signal now propagates to child workflow execution
  • Workflow Node Limit — Max 500 nodes per workflow (DoS prevention)
  • Workflow Copilot Prompt — All 23 node types documented with correct short names matching convertDefinitionToReactFlow
  • Workflow Wizard Templates — Rewritten with valid node types (previously used non-existent start/ai/end types)
  • Merge Node firstCompleted — Mode parameter now functional (was no-op returning same as waitAll)
  • Notification Node — Awaits broadcast instead of fire-and-forget, surfaces delivery warnings
  • Code Node — Validates language input (rejects unsupported languages instead of silent fallback)
  • Delay Node — Logs warning when 1-hour safety cap is applied
  • Agent Concurrent Guard — cycleInProgress flag prevents double cycle execution
  • Agent Rate Limit Retry — Re-schedules with backoff after throttling (was silently stopping)
  • Agent Crew Context Cache — 30-second TTL cache reduces N*3 DB queries per heartbeat to 1
  • getCommunicationBus() — Throws descriptive error instead of unsafe non-null assertion crash
  • Subagent spawnCounts — Cleaned up when conversations have no active sessions (was growing unbounded)
  • Workspace Creation — Warning logged instead of silent debug on failure

Changed

  • Extracted agent-runner-utils.ts — shared tool registration, agent factory, model resolution, timeout, JSON parsing (~360 LOC dedup)
  • Centralized scheduling constants (config/defaults.ts) — MANAGER_MAX_CONSECUTIVE_ERRORS, MANAGER_SESSION_PERSIST_INTERVAL_MS, per-system delay bounds
  • AutonomousAgentResult base interface for unified result types across all runners
  • Fleet mission context included in all worker types (coding-cli, mcp-bridge were missing it)

Security

  • Bump hono 4.12.3 → 4.12.8 (arbitrary file access, prototype pollution, cookie/SSE injection)
  • Bump @hono/node-server 1.19.9 → 1.19.11 (authorization bypass via encoded slashes)
  • Bump undici override >=6.23.0 → >=6.24.1 (WebSocket DoS, CRLF injection, request smuggling)

Testing

  • 26,650+ tests total (core: 9,832; gateway: 16,236; ui: 141; cli: 293; channels: 148)
  • New: 68 Fleet Command tests, workflow node executor improvements

0.1.8 - 2026-03-14

Added

  • Unified Channel System — Extensible channel SDK with UnifiedBus event router and UCP (Unified Channel Protocol) adapters; channels register via builder pattern with standardized lifecycle hooks
  • Web Chat Channel Plugin — Embeddable floating chat widget for websites with real-time WebSocket messaging, session management, and customizable appearance
  • SMS Channel Plugin — SMS messaging channel via Twilio integration with send/receive support
  • Email Channel Plugin — Email channel via SMTP/IMAP with inbound parsing and outbound delivery
  • Matrix Channel Plugin — Matrix protocol channel for federated messaging support
  • Cross-Channel Notification System — Unified notification dispatching across all connected channels with priority routing, delivery tracking, and fallback chains
  • Fleet System — Multi-worker fleet management with worker assignment, budget configuration, schedule parameters, real-time status events, and full admin UI with detail panels
  • Claw Mode (Autonomy L5) — Enhanced crew orchestration mode with elevated autonomy capabilities for coordinated multi-agent operations
  • Crew UI Panels — Shared memory and task queue panels for crew collaboration visibility
  • Dashboard Enhancements — Claw badges and crew metrics widgets on the main dashboard
  • ACP for Coding Agents — Agent Communication Protocol enabled for all coding agent providers with critical bug fixes
  • Autonomy Logging — Comprehensive structured logging across all autonomous systems (pulse, heartbeat, triggers, background agents)
  • Database Backup Endpoints — Backup listing and download via REST API
  • Setup Wizard Documentation — Interactive setup wizard docs with type safety improvements

Fixed

  • Fleet Boot/Lifecycle — Fixed critical bugs preventing fleet system from starting: worker config snake_case→camelCase mapping, race condition guard, missing tool registrations
  • Autonomous Agent Scheduling — Fixed provider fallback, duration calculation, and scheduling bugs in background agents
  • Session Shutdown Cleanup — Removed hardcoded limit:100 truncation and added proper session cleanup on shutdown
  • Production Hardening — Fixed shutdown ordering, memory leaks in timers, security edge cases, and logging gaps
  • Autonomy Performance — Skip LLM call when no signals detected; skip high-frequency tool events in trigger engine
  • Silent Failures — Upgraded swallowed errors to warn-level logging across autonomous systems
  • CLI Provider Endpoints — Added CLI provider support to detail and models API endpoints
  • Docker Monorepo Build — Fixed Dockerfile for correct monorepo structure
  • UI Wizard Scroll — Fixed wizard buttons scrolling out of view
  • Husky Pre-Commit Hook — Added missing shebang line
  • WebSocket Server Test — Fixed circular import TDZ error and missing mock exports

Changed

  • Comprehensive complexity reduction refactoring across the codebase
  • Database API path changed from /database to /db
  • Removed unused dependencies and dead code cleanup

Performance

  • Skip LLM call when no autonomy signals detected
  • Skip high-frequency tool events in trigger engine evaluation

0.1.6 - 2026-03-06

Added

  • Conversation Sidebar — Persistent conversation sidebar with ID-based session persistence, inline rename support, and auto-refresh on channel events
  • Pairing Key Ownership — Per-channel rotating pairing keys for channel access control with revoke support; pairing key banner on Channels page
  • WhatsApp Group Support — Group message storage, group messages API endpoint, group/chat listing endpoints, and passive history sync with on-demand fetch
  • WhatsApp Anti-Ban Hardening — P0 anti-ban safety filters, auto-reply protection, history sync race condition fixes, and 440 connectionReplaced reconnect loop fix
  • WhatsApp Auto-Claim Owner — Automatically claim channel ownership on first self-chat message, removing the need for manual /connect
  • Crew Orchestration Engine — Runtime crew orchestration with Plans tab integration and agent lifecycle fixes
  • Debug System Prompt Breakdown — Full system prompt section breakdown logged on every request with DebugDrawer UI showing per-section drill-down
  • Perplexity Agent API — Added Perplexity Agent API provider configuration

Fixed

  • Anthropic Prompt Caching — Moved orchestrator to static cache block; round time context to hour boundary to prevent cache invalidation; moved extensions before cache split point
  • Provider Empty Content — Send null content instead of "" for assistant messages with tool_calls (fixes Anthropic validation errors)
  • Composio Integration — Fixed getAvailableApps flat-array response handling, meta field mapping, and invalid 'google' slug
  • Chat Double-Persistence — Extracted ConversationService, fixed messages being persisted twice
  • WhatsApp QR Code — Fixed QR code not appearing when session is stale/expired
  • WhatsApp Self-Chat — Fixed intermittent self-chat send failures
  • Channel Owner Flow — Reply with /connect instructions when no owner claimed; sidebar auto-refresh on channel events
  • UI Logout Button — Show Logout button in both connected and disconnected channel states
  • Extension Creator DOM Error — Resolved insertBefore DOM error in extension creator (#8)
  • Debug Log Truncation — Removed truncation from debug log entries, store full section content
  • Soul Agent Pipeline — 17 FIXPLAN fixes across heartbeat pipeline and extension creator (commit 12e996d)

Changed

  • Extracted ConversationService from chat route for cleaner separation of concerns
  • Channel history improvements for better message threading
  • Removed unused code and improved type safety in resource tools
  • Removed generated models-dev-full.json artifact, updated .gitignore

Security

  • SSRF / DNS Rebinding Protection — isBlockedUrl() sync hostname check + isPrivateUrlAsync() with DNS rebinding detection and 1-min cache; applied to browser service, /fetch-url, and web-fetch executors
  • Rate Limiter TTL Cleanup — Fixed memory leak in sliding window rate limiter
  • XSS Escaping — Added output escaping for user-controlled content in HTML responses

Testing

  • 389+ test files, 22,100+ tests total
  • New: soul agent unit tests (127 core tests across 6 files), gateway soul coverage (souls repo, communication tools, heartbeat service)

0.1.5 - 2026-03-02

Added

  • Soul Agent System — Rich agent identity framework with personality, mission, role, relationships, heartbeat-driven lifecycle, evolution tracking, and boot sequences; gateway service with full CRUD and heartbeat execution
  • Autonomous Hub — Unified command center consolidating soul agents, background agents, crews, messaging, and activity into a single tabbed dashboard with search, filters, and real-time WebSocket status
  • AI Agent Creator — Conversational agent creation via SSE streaming chat with a dedicated designer agent; describe what you need in plain language, refine through conversation, preview the JSON config, and create in one click
  • Crew System — Multi-agent crews with role assignments, delegation protocols, and crew templates for coordinated multi-agent workflows
  • Agent Communication — Inter-agent messaging system with inbox, compose, and message history; CommsPanel for viewing and sending messages between agents
  • Activity Feed — Unified timeline of heartbeat logs and agent messages with aggregate stats (total runs, success rate, avg duration, total cost)
  • 16+ Agent Templates — Pre-built agent configurations (Morning Briefer, News Monitor, Code Reviewer, Budget Tracker, Social Media Manager, Health & Wellness Coach, etc.) with one-click creation
  • Agent Profile Page — Detailed agent view with identity display, inline soul editor, heartbeat configuration, action controls (pause/resume/delete), and error/not-found handling
  • Global Status Bar — Compact header showing live agent count, running/paused/error breakdown, daily cost, WebSocket connection state, and autonomy settings link
  • 77 new AI providers — Updated provider model data with 77 providers and 1 new addition

Fixed

  • AI Creator chatbot behavior — Fixed AI Agent Creator acting like a regular chatbot instead of designing agent configs; root cause was BASE_SYSTEM_PROMPT overriding the inline designer instruction; now uses a dedicated __ai_agent_designer agent with proper system prompt via agentId routing
  • Autonomous Hub modal backgrounds — Replaced invalid bg-surface with valid bg-bg-primary Tailwind token
  • WebSocket connection status — Fixed useAgentStatus hook to properly return isConnected state for live/offline indicator
  • Empty state UX — Added differentiated empty state buttons (Browse Templates / Chat with AI / Create Manually) with separate wizardInitialStep tracking
  • Agent profile error handling — Distinguished between loading errors (with retry) and agent-not-found (with navigation back) in AgentProfilePage
  • Activity feed error state — Added inline error display with retry button in ActivityFeed
  • Template fetch errors — Added toast notification when crew template loading fails on hub mount
  • CommsPanel improvements — Fixed "from" field population in compose form and added loading spinner during data fetch
  • TemplateCatalog search — Added clear button to search input for better UX
  • Silent catches — Replaced empty catch blocks with user-facing error toasts in CrewSection, CommsPanel, and SoulEditor
  • Accessibility — Added aria-label to all icon-only buttons across the autonomous hub (back button, stop generating, tag remove, close creator)
  • Type safety — Removed unsafe as any and as unknown type assertions across all autonomous hub components
  • Structured logging — Converted template-literal error logs to structured context objects in gateway observability
  • Fetch security — Added fetch timeout and download size limits with budget enforcement tests
  • 12 bugs from BUGS.md — Resolved P0 through P3 priority bugs plus SEC-001 security finding

Changed

  • Provider model data updated (77 providers)
  • Soul tools registered in gateway providers with formatted structured log context
  • Cleaned up unused imports, dead code, and repo artifacts

Testing

  • 389+ test files, 22,100+ tests total

0.1.4 - 2026-02-28

Added

  • Background Agents — Persistent autonomous agents that run independently on interval, continuous, or event-driven schedules with rate limiting, budget tracking, auto-pause on errors, and graceful shutdown
  • Background Agent Full Tool Access — Background agents now have the same capabilities as chat agents: all 170+ tools, extension/skill tools, plugin tools, MCP tools, memory injection, and configurable provider/model selection
  • Background Agent Workspace Isolation — Each background agent gets an isolated file workspace for safe file operations
  • WhatsApp Baileys Integration — Replaced Meta Cloud API with Baileys library for WhatsApp; QR code authentication (no Meta Business account needed), self-chat mode with loop prevention, session persistence in app data directory
  • Channel User Approval System — Multi-step verification for channel users: approval code flow, manual admin approval, user blocking/unblocking with real-time notifications
  • EventBus Deep Integration — Unified event backbone across the entire system; EventBusBridge translates dot-notation events to WebSocket colon-notation for real-time UI updates
  • Event Monitor UI — Live event stream viewer for debugging EventBus events in the web UI
  • Extension SDK — Extensions can call any of 150+ built-in tools via utils.callTool(), with utils.listTools(), Config Center access, and blocked tool enforcement
  • 6 Default Extensions — Daily Briefing, Knowledge Base, Project Tracker, Smart Search, Automation Builder, Contact Enricher bundled out-of-the-box
  • Extension Security Audit — LLM-powered security analysis for skills and extensions before installation
  • Selective Extension Injection — Request-preprocessor routing for targeted extension injection per conversation
  • Channel Soft Disconnect / Hard Logout — disconnect() preserves session for instant reconnect; logout() clears session data requiring re-authentication (e.g. new QR scan)
  • Workflow Enhancements — 7 new node types, input_schema column, workflow versioning and approvals

Changed

  • Extension tools synced into shared ToolRegistry with ext.*/skill.* namespace prefixes
  • Channel user events (first_seen, blocked, unblocked, pending) emitted via EventBus with complete WS forwarding
  • Channels reduced to Telegram + WhatsApp (Discord/Slack/LINE/Matrix removed)

Fixed

  • Scheduler Day Boundary — getNextRunTime test failed on month-end dates (e.g. Feb 28 → Mar 1) due to incorrect rollover arithmetic
  • Vitest Constructor Mocks — Fixed test stability issues with constructor mocking patterns across gateway tests
  • Test Helpers Build Error — Added explicit return types to test-helpers to fix TS2742 build error

Testing

  • 366+ test files, 21,500+ tests total
  • New: background-agent-manager, background-agent-runner, background-agent-tools, service-impl logout tests

0.1.3 - 2026-02-26

Added

  • Model Routing — Per-process model selection (chat, telegram, pulse) with provider fallback chains, configurable via API and UI
  • Extended Thinking — Anthropic extended thinking support with configurable budget tokens for deeper reasoning
  • Sidebar Reorganization — Navigation menus reordered by usage frequency: daily items at top, power-user features in collapsible groups, settings ordered by domain

Fixed

  • Telegram FK Constraint — Second message after server restart failed with channel_sessions_conversation_id_fkey violation; conversation recovery now persists to DB before updating FK (fixes #7)
  • Dashboard Streaming — Null model parameter caused TypeScript build failures in generateAIBriefingStreaming
  • Expenses Page — Feb-31 date bug when filtering by month; added edit support with modal form
  • SystemPage Polling — Database operation status polling leaked timers on unmount; added ref-based cleanup
  • ApiKeysPage — Default model save silently swallowed errors; now shows toast feedback
  • AutonomyPage — Promise.all → Promise.allSettled so partial API failures don't blank the page
  • ModelsPage — Settings link pointed to /settings instead of /settings/api-keys
  • WorkspacesPage — Empty workspace badge showed for workspaces with 1 file (should be 0)
  • CalendarPage — No validation when end date/time was before start
  • TasksPage — Missing cancelled status in filter and visual styling
  • SKILL.md Parser — Improved YAML metadata parsing for block sequences and nested maps

Changed

  • Config Center: removed 5 orphaned seed services (Deepgram, DeepL, Tavily, Serper, Perplexity) with no built-in consumer code
  • Gateway routes: parseJsonBody helper adopted across all route modules
  • Dev dependencies bumped: ESLint 10.0.2, Turbo 2.8.11, typescript-eslint 8.56.1
  • .gitignore: broader protection patterns for stray generated files

0.1.2 - 2026-02-26

Added

  • CLI Tools Platform — 40+ discoverable CLI tools with automatic PATH-based binary detection, categorization (linters, formatters, build tools, package managers, security scanners, databases, containers), and version detection
  • Per-Tool Security Policies — allowed (auto-execute), prompt (require approval), blocked (reject) per user per tool, with batch policy updates via API
  • Dynamic Risk Scoring — Catalog-based risk levels (low/medium/high/critical) feed into the autonomy risk engine, overriding generic tool risk scores
  • Custom CLI Tool Registration — Register any binary as a CLI tool with category and risk metadata via POST /cli-tools/custom
  • CLI Policy Approval Integration — Per-tool policies wired into the real-time approval flow in the orchestrator, dynamic risk scoring based on catalog risk levels
  • Coding Agents — Orchestrate external AI coding CLIs (Claude Code, Codex, Gemini CLI) with session management, real-time terminal output streaming, and result persistence
  • Dual Execution Modes — Auto mode (headless child_process.spawn) and interactive mode (PTY terminal) for coding agents
  • Custom Coding Agent Providers — Register any CLI binary as a coding agent provider via the CLI Providers API
  • Model Routing — Per-process model selection (chat, telegram, pulse) with fallback chains, configurable via API and UI
  • Extended Thinking — Anthropic extended thinking support for deeper reasoning in complex tasks

Changed

  • Gateway route modules: 40 → 43 top-level (added coding-agents.ts, cli-tools.ts, cli-providers.ts, model-routing.ts)
  • Repositories: 37 → 41 (added coding-agent-results, cli-providers, cli-tool-policies, autonomy-log)
  • UI pages: 41 → 47 (added CodingAgentsPage, CodingAgentSettingsPage, CliToolsSettingsPage, ModelRoutingPage, SecurityPage, AboutPage)
  • WebSocket events: added coding-agent:session:* for coding agent lifecycle

Testing

  • 315+ test files, 19,200+ tests total
  • New: coding-agent-service, coding-agent-sessions, cli-providers, cli-tool-policies, coding-agent-results repository tests

0.1.1 - 2026-02-23

Added

  • Pulse System — Autonomous AI-driven engine that proactively gathers context, evaluates signals, invokes the LLM, executes actions, and reports results on an adaptive timer (5–15 min)
  • Pulse Directives — Configurable evaluation rules, action cooldowns, blocked actions, custom instructions, and 4 preset templates (Balanced, Conservative, Proactive, Minimal)
  • Pulse Execution Lock — Prevents concurrent pulse execution; manual and auto pulses share the same lock
  • Pulse Activity Broadcasting — Real-time WebSocket pulse:activity events with stage progression (starting → gathering → evaluating → deciding → executing → reporting → done)
  • Pulse Activity Monitor (UI) — Live activity banner with stage name and elapsed time, "Run Now" button disables during pulse, 409 toast on concurrent attempts
  • Pulse History & Stats — Paginated pulse log with signal IDs, urgency scores, action results, and expandable details
  • Pulse Route Guard — POST /pulse/run returns 409 ALREADY_RUNNING when a pulse is in progress

Changed

  • AutonomyEngine.getStatus() now includes activePulse field (null when idle)
  • Broadcaster in server.ts routes pulse:activity events separately from system:notification

Testing

  • 315 test files, 19,100+ tests total
  • New: 5 engine execution lock tests + 2 route guard tests

0.1.0 - 2026-02-22

Initial release of OwnPilot.

Added

  • Multi-Provider AI — 4 native providers (OpenAI, Anthropic, Google, Zhipu) + 8 aggregators (Together AI, Groq, Fireworks, DeepInfra, OpenRouter, Perplexity, Cerebras, fal.ai) + any OpenAI-compatible endpoint
  • Local AI Support — Auto-discovery for Ollama, LM Studio, LocalAI, and vLLM
  • Smart Provider Routing — Cheapest, fastest, smartest, balanced, and fallback strategies
  • Anthropic Prompt Caching — Static system prompt caching to reduce input token costs
  • Context Management — Real-time token tracking, detail breakdown, and AI-powered context compaction
  • 170+ Built-in Tools across 28 categories (personal data, files, code execution, web, email, media, git, translation, weather, finance, automation, vector search, data extraction, utilities)
  • Meta-tool Proxy — Only 4 meta-tools sent to the LLM; all tools available via dynamic discovery
  • Tool Namespaces — Qualified tool names (core., custom., plugin., skill., mcp.)
  • MCP Integration — Client (connect to external MCP servers) and Server (expose tools to MCP clients)
  • User Extensions — Installable tool bundles with custom tools, triggers, services, and configs
  • Skills — Open standard SKILL.md format (AgentSkills.io) for instruction-based AI knowledge packages
  • Custom Tools — Create new tools at runtime via LLM (sandboxed JavaScript)
  • Connected Apps — 1000+ OAuth integrations via Composio
  • Personal Data — Notes, Tasks, Bookmarks, Contacts, Calendar, Expenses with full CRUD
  • Productivity — Pomodoro timer, habit tracker, quick capture inbox
  • Memories — Long-term persistent memory with importance scoring, vector search, AES-256-GCM encryption
  • Goals — Goal creation, decomposition, progress tracking, next-action recommendations
  • Custom Data Tables — User-defined structured data with AI-determined schemas
  • 5 Autonomy Levels — Manual, Assisted, Supervised, Autonomous, Full
  • Triggers — Schedule-based (cron), event-driven, condition-based, webhook
  • Heartbeats — Natural language to cron conversion for periodic tasks
  • Plans — Multi-step autonomous execution with checkpoints and retry logic
  • Workflows — Visual multi-step automation with drag-and-drop builder and Workflow Copilot
  • Web UI — React 19 + Vite 7 + Tailwind CSS 4 with 41 pages, 60+ components, dark mode
  • Telegram Bot — Grammy-based bot with user/chat filtering and message splitting
  • WebSocket — Real-time broadcasts for all data mutations
  • REST API — 40 route modules with standardized responses, pagination, and error codes
  • Sandboxed Code Execution — Docker isolation, VM, Worker threads with 4-layer security
  • PII Detection & Redaction — 15+ categories
  • Zero-Dependency Crypto — AES-256-GCM + PBKDF2 using only Node.js built-ins
  • Authentication — None, API Key, or JWT modes
  • Rate Limiting — Sliding window with burst support
  • Tamper-Evident Audit — Hash chain verification for audit logs

Infrastructure

  • TypeScript 5.9 monorepo with Turborepo
  • 307 test files with 19,200+ tests (Vitest)
  • GitHub Actions CI/CD pipeline
  • Docker multi-arch image (amd64 + arm64) published to ghcr.io/ownpilot/ownpilot
  • PostgreSQL with pgvector for vector search