# Andres Echeverria — Full LLM Context > Complete machine-readable portfolio context for https://estebanech.com. Prefer [/llms.txt](https://estebanech.com/llms.txt) for navigation; use this file when a single fetch is enough. Generated from the live site content sources. --- # Andres Echeverria | Software & AI Engineering > Software engineering student and independent software developer focused on full-stack systems, applied AI, automation, and developer tools. I build practical software across mobile, web, backend, and local AI environments, with an emphasis on reliability, privacy, and real-world use. This is the machine-readable home page for https://estebanech.com. The interactive UI is a terminal portfolio; prefer this file, [/llms.txt](https://estebanech.com/llms.txt), or [/llms-full.txt](https://estebanech.com/llms-full.txt) for reliable extraction. ## Quick facts - Name: Andres Echeverria Giraldo (estebanech) - Location: San Jose, California - Focus: Software engineering; AI engineering; Full-stack product development; Local and agentic AI systems - Email: estebanech16@gmail.com - GitHub: https://github.com/Estebanech1620 - DestinyX GitHub: https://github.com/DestinyX-Studios-LLC - LinkedIn: https://www.linkedin.com/in/andresecheverriag ## Primary links - [About](https://estebanech.com/about.md): profile, education, skills - [Projects](https://estebanech.com/projects.md): catalog of featured work - [LOLE OS](https://estebanech.com/projects/lole-os.md): completed client commerce OS case study - [Forge](https://estebanech.com/projects/forge.md): local AI coding agent case study - [Saki](https://estebanech.com/projects/saki.md): permission-aware Discord agent case study - [DestinyX Studios](https://estebanech.com/projects/destinyx.md): product company with selective partner engineering — Oppli, Forge, and private systems - [VidVortex](https://estebanech.com/projects/vidvortex.md): open-source media downloader - [Oppli](https://www.oppli.app): DestinyX career intelligence product (no portfolio case study) - [llms.txt](https://estebanech.com/llms.txt): curated LLM index - [llms-full.txt](https://estebanech.com/llms-full.txt): full concatenated site context ## Current positioning Lead Software Engineer & Founder at DestinyX Studios (2024 – Present). Building Oppli and Forge as DestinyX products, with selective partner engineering for private systems; recent freelance client delivery includes LOLE OS for LOLE Boutique. --- # About Andres Echeverria > Software engineering student and independent software developer focused on full-stack systems, applied AI, automation, and developer tools. I build practical software across mobile, web, backend, and local AI environments, with an emphasis on reliability, privacy, and real-world use. - HTML: https://estebanech.com/about - Markdown: https://estebanech.com/about.md - Portfolio: https://estebanech.com - LLM index: https://estebanech.com/llms.txt ## Profile - Name: Andres Echeverria Giraldo (estebanech) - Location: San Jose, California - Focus: Software engineering, AI engineering, Full-stack product development, Local and agentic AI systems - Email: estebanech16@gmail.com - GitHub: https://github.com/Estebanech1620 - DestinyX GitHub: https://github.com/DestinyX-Studios-LLC - LinkedIn: https://www.linkedin.com/in/andresecheverriag ## Experience (current) - Lead Software Engineer & Founder — DestinyX Studios — 2024 – Present (San Francisco Bay Area). Software product and AI engineering company with a selective partner-engineering practice. Owned products: Oppli and Forge. Partners: private platforms, agents, automation, and custom infrastructure. Long-term direction: agent infrastructure, model specialization, and original AI systems when evidence justifies it. Case study: https://estebanech.com/projects/destinyx ## Experience (contract) - Freelance Software Engineer — LOLE Boutique — July 2026 (Remote, Independent Contractor). Completed client engagement delivering LOLE OS, a staff-only internal commerce operating system (Shopify mirrors, Supabase auth/RLS, Railway workers, controlled OpenAI workflows). Sole engineer. Repository private. Not employment at LOLE Boutique. Case study: https://estebanech.com/projects/lole-os ## Experience (Apple) - Lab Engineer | Machine Learning Platform Technologies — May 2026 – Jul 2026 (Cupertino, CA) - RF Test Engineer — NPI Validation — Feb 2024 – Feb 2026 - Lab Lead | Early Field Failure Analysis — Aug 2023 – Feb 2024 - Mac Genius — Mar 2019 – Mar 2023 - Technical Specialist — Feb 2020 – Apr 2021 - AppleCare Technical Advisor — Apr 2020 – Jun 2020 ## Experience (other) - Computer Science Summer Institute — Google — Jul 2020 – Aug 2020 (Mountain View, CA) ## Education - Western Governors University — B.S. Software Engineering (expected Dec 2026) - Western Governors University — M.S. AI Engineering (expected May 2027) ## Certifications - ACMT - ACiT - Python Scripting for Automation - Bash Scripting & Shell Programming ## Skills ### Languages - TypeScript - JavaScript - Python - SQL - Bash ### Application Engineering - React - React Native - Next.js - Expo - Node.js - Express - Server Actions - REST APIs - GraphQL - webhooks - OAuth ### AI Systems & Agents - Tool-calling agents - RAG - structured outputs - model routing - Context management - semantic memory - multimodal workflows - OpenAI - Groq - Ollama - Qwen - MCP - Vercel AI SDK ### Data & Persistence - PostgreSQL - Supabase - SQLite - FTS5 - LanceDB - Drizzle ORM - pgvector - Redis - BullMQ - Row-Level Security ### Infrastructure & Integrations - Vercel - Railway - Docker - GitHub Actions - EAS Build - Shopify Admin API - Discord Gateway - Lavalink - background workers ### Quality, Security & Operations - CI/CD - automated testing - debugging - observability - Role-based access control - audit logging - feature flags - Software validation - hardware-software integration - Python automation ## Philosophy - Ship real software, not demos - Think in systems, not features - Explain tradeoffs, don't hide them - Learn from failures, celebrate wins - Build for scale, optimize when needed --- # Projects — estebanech > Featured and shipped work by Andres Echeverria: client platforms, local AI agents, studio products, and developer tools. - HTML: https://estebanech.com/projects - Markdown: https://estebanech.com/projects.md - LLM index: https://estebanech.com/llms.txt - Full context: https://estebanech.com/llms-full.txt ## Catalog ### DestinyX Studios DestinyX Studios builds customer-facing products of its own and engineers private software systems for select partners, with growing specialization in agent infrastructure, developer tools, and AI-enabled applications. Current products: Oppli and Forge. - Label: SOFTWARE PRODUCT & AI ENGINEERING COMPANY - Type: Software Product & AI Engineering Company - Role: Lead Software Engineer & Founder - Status: Active - Repository: Studio (https://github.com/DestinyX-Studios-LLC) - Stack: Owned products, Partner engineering, Agent systems, Private platforms, Developer tools - Tags: Company, AI Engineering, Full-Stack, Developer Tools, Partner Engineering - Page: https://estebanech.com/projects/destinyx - Markdown: https://estebanech.com/projects/destinyx.md ### Oppli AI career coach for your next role — helps prioritize which jobs are worth your time, what is holding you back, and what to fix first. React Native app, Express API, and Next.js admin with ranking, embeddings, and LLM workflows. DestinyX Studios product (Beta). - Label: AI CAREER COACH - Type: DestinyX Studios Product - Role: Founder & Sole Engineer - Status: Active Development - Repository: Private - Stack: React Native, TypeScript, Expo, Express, Next.js, PostgreSQL, Redis, BullMQ, OpenAI - Tags: Full-Stack, AI Engineering, Career Tech, Mobile - Page: https://www.oppli.app ### LOLE OS A secure staff-only commerce and operations platform for LOLE Boutique — Shopify mirrors, inventory, customers, contribution reporting, AI-assisted workflows, and platform administration. Paid independent contractor engagement. - Label: INTERNAL COMMERCE OPERATING SYSTEM - Type: Client Engagement - Role: Freelance Software Engineer - Status: Completed Client Engagement - Repository: Private - Stack: Next.js 16, React 19, TypeScript, Supabase, PostgreSQL, Shopify GraphQL, Railway, OpenAI - Tags: Client Work, Full-Stack, Commerce Infrastructure, Platform Engineering - Page: https://estebanech.com/projects/lole-os - Markdown: https://estebanech.com/projects/lole-os.md ### Forge A privacy-first, on-device coding agent that combines structured tool execution, project RAG, persistent memory, human approval gates, MCP extensions, context management, and optional model specialization on Apple Silicon. - Label: LOCAL AI CODING AGENT - Type: DestinyX Studios Product - Role: Sole Designer and Engineer - Status: Active Development - Repository: Private - Stack: Python, Ollama, Qwen, SQLite, LanceDB, MCP, MLX, prompt_toolkit - Tags: AI Engineering, Developer Tools, Local LLM, Agentic Systems - Page: https://estebanech.com/projects/forge - Markdown: https://estebanech.com/projects/forge.md ### Saki A private Discord operations assistant combining a permission-aware LLM tool agent, deterministic moderation, persistent SQLite memory, scheduled workflows, multimodal input, and Lavalink-powered music. - Label: AGENTIC DISCORD SERVER ASSISTANT - Type: Independent Production Project - Role: Sole Designer and Engineer - Status: Live - Repository: Private - Stack: TypeScript, Node.js, discord.js, Groq, SQLite, FTS5, Lavalink, Docker, Railway - Tags: AI Engineering, Backend Systems, Discord, Agentic Systems - Page: https://estebanech.com/projects/saki - Markdown: https://estebanech.com/projects/saki.md ### VidVortex A desktop media downloader with quality picking, per-OS setup scripts, and GitHub Releases packaging — paste a URL, choose video or audio, and download without juggling CLI flags. - Label: CROSS-PLATFORM MEDIA DOWNLOADER - Type: Open Source Tool - Role: Sole Designer and Engineer - Status: Shipped - Repository: Public (https://github.com/Estebanech1620/VidVortex) - Stack: Python, yt-dlp, ffmpeg, PyInstaller, GitHub Actions - Tags: Developer Tools, Desktop, Packaging - Page: https://estebanech.com/projects/vidvortex - Markdown: https://estebanech.com/projects/vidvortex.md --- # LOLE OS > LOLE OS is a staff-only internal operating system for LOLE Boutique — Shopify commerce mirrors, inventory, customers, contribution reporting, AI-assisted workflows, and platform administration. Built as independent contractor client work by Andres Echeverria. - HTML: https://estebanech.com/projects/lole-os - Markdown: https://estebanech.com/projects/lole-os.md ## Building an Internal Commerce Operating System for LOLE Boutique LOLE OS is a secure staff-only command center that brings Shopify commerce operations, inventory, customers, contribution reporting, AI-assisted workflows, and platform administration into one application. It is designed around a clear separation of responsibility: Shopify owns commerce truth, Supabase owns internal state and authorization, and LOLE OS owns staff workflows and business interpretation. ## Metadata - Role: Freelance Software Engineer - Client: LOLE Boutique - Type: Independent Contractor - Status: Completed Client Engagement - Dates: July 2026 - Location: Remote - Version: lole-os@0.1.0 - Delivered scope: Feature Freeze v1.0 - Repository: Private ## Overview LOLE OS is a production-oriented internal business operating system built for LOLE Boutique staff. It is not a storefront, Shopify theme, Shopify Admin clone, or analytics dashboard — it is the operational layer where authorized staff manage commerce mirrors, inventory interpretation, customers, contribution reporting, platform health, and controlled AI workflows. The platform was designed and developed as paid freelance independent contractor work. Platform access and ownership belong to the client. Andres Echeverria delivered the engineering as sole engineer: architecture, frontend, backend, database design, integrations, deployment planning, and technical documentation. ## Problem Boutique retail operations often fragment across Shopify Admin, spreadsheets, social platforms, invoice inboxes, and ad-hoc messaging. Staff need a secure internal surface that interprets business reality without replacing Shopify as the commerce source of truth — and without exposing secrets or inventing financial KPIs through unconstrained AI. The engineering question behind LOLE OS was: > How do you build a staff-only operating system that mirrors Shopify commerce, enforces real authorization, keeps numbers deterministic, and adds AI only where extraction and interpretation help — without becoming a shadow storefront or an ungoverned chatbot? ## Solution ### Staff operating surfaces Operational views for daily alerts, store management, inventory, contribution reporting, clients, collections, marketing, staff access, strategy, and platform settings — bilingual English/Spanish where staff workflows require it. ### Dual-source architecture Shopify remains authoritative for products, inventory, orders, and customers. LOLE OS mirrors commerce data and adds internal operational context in Supabase without attempting to replace Shopify's responsibilities. ### Authorization and audit Supabase Auth, PostgreSQL row-level security, server-side capability checks, role assignment RPCs, audit trails, and feature flags keep access and platform controls enforceable outside the browser. ### Synchronization and workers Shopify Admin GraphQL sync with signed webhooks, deduplicated delivery processing, scheduled reconciliation, and a Railway background worker for products, orders, customers, Instagram analytics, reports, AI briefs, and platform health. ### Controlled AI subsystem OpenAI powers grounded staff assistance, multimodal invoice intake with human review, and daily operational briefs based on precomputed metrics. Models extract or explain; they do not invent business KPIs. ## Capabilities ### Commerce operations Staff workflows over mirrored Shopify products, inventory, orders, customers, fulfillment context, and contribution reporting. ### Role-based access Staff roles including owner, engineer, developer, admin, and pending/customer resolved server-side with RLS and capability checks. ### Developer control portal Platform health, integrations, jobs, webhooks, AI usage, feature flags, audit trails, and environment-presence diagnostics. ### Bilingual workflows English and Spanish support via next-intl, stored locale preferences, and business-specific Latin American Spanish terminology. ### Invoice intake with review Multimodal invoice extraction with staff review, inventory application, cost authorization, failure states, and reversible snapshots. ### Grounded operations assistant A read-oriented staff assistant limited to a small number of tool rounds, backed by business tools rather than unconstrained freeform tool use. ### Daily operational briefs Automated briefs based on precomputed metrics so models explain and summarize rather than invent KPIs. ### Shopify freshness Signed webhooks plus scheduled reconciliation keep mirrors current while deduplicating delivery processing. ## Principles ### Shopify remains the commerce source of truth. LOLE OS mirrors commerce data and adds internal operational context without attempting to replace Shopify's responsibility for products, inventory, orders, and customers. ### Supabase owns internal state. Supabase stores authentication, staff profiles, role data, operational mirrors, alerts, reports, audit history, assistant threads, AI usage, feature flags, and invoice workflows. ### Server-side authorization. Roles are resolved through authenticated server state. Client-supplied roles are never trusted. ### Deterministic numbers before AI. Contribution, shipping absorption, stock logic, VIP classification, and operational metrics are computed in code. AI is used for extraction, summarization, and interpretation. ### Human approval for AI actions. Invoice extraction does not directly mutate inventory without staff review and authorization. Invoice applications support reversible snapshots. ### Feature flags as platform controls. AI, synchronization, reporting, commerce writes, and health checks can be disabled globally without redeploying the application. Kill switches are enforced in both UI and API routes. ### Secrets remain server-side. Client interfaces may show whether an integration is configured, but never reveal secret values such as tokens or API keys. ## Highlights - Product version: lole-os@0.1.0 - Delivered scope: Feature Freeze v1.0 - Supported languages: English and Spanish - Staff roles: owner, engineer, developer, admin, pending/customer - AI assistant limited to a maximum of three tool rounds - Shopify freshness via webhooks plus scheduled reconciliation - Background processing on a separate Railway worker - Platform-wide AI kill switches in UI and API routes - Invoice applications support reversible snapshots - Dedicated developer control portal ## Technology ### Core application - Next.js 16 - React 19 - TypeScript - Node.js - Tailwind CSS - Zod ### Infrastructure and persistence - Supabase Auth - PostgreSQL - Row-Level Security - Supabase Storage - Vercel - Railway - Docker ### Integrations - Shopify Admin GraphQL - Shopify Webhooks - OpenAI - Vercel AI SDK - Google OAuth - Instagram Graph API ### Product capabilities - Server Actions - Background Workers - Webhooks - Feature Flags - Audit Logging - Internationalization - Role-Based Access Control ## Status LOLE OS was delivered as a freelance independent contractor engagement in July 2026 at Feature Freeze v1.0 (lole-os@0.1.0). The production repository remains private because it contains client configuration conventions, integration credentials handling, operational workflows, and internal business logic. A public case study describes architecture and engineering responsibility without exposing secrets, customer data, or confidential client strategy. Built by Andres Echeverria for LOLE Boutique as freelance independent contractor work. Platform access and ownership belong to the client. This is not employment at LOLE Boutique. ## Next Steps - Expand messaging integrations and operational alerts under existing authorization and feature-flag controls - Continue hardening reconciliation and webhook failure diagnostics - Deepen contribution and inventory reporting views using deterministic metric pipelines - Improve operator documentation for bilingual staff workflows - Iterate on AI brief quality while preserving deterministic KPI computation Author: Andres Echeverria (estebanech) --- # Forge > Forge is a privacy-first local AI coding agent for Apple Silicon with structured tools, project RAG, persistent memory, human approval gates, MCP extensions, and optional model specialization. - HTML: https://estebanech.com/projects/forge - Markdown: https://estebanech.com/projects/forge.md ## Building a Local AI Coding Agent for Apple Silicon Forge is a private local agent runtime that lets open-weight models inspect repositories, modify code, run commands and tests, manage Git workflows, perform research, use persistent memory, and connect to external MCP tools without sending the core coding workflow to a cloud model provider. ## Metadata - Role: Sole Designer and Engineer - Type: DestinyX Studios Product - Status: Active Development - Version: 0.1.0 - Platform: Apple Silicon - Repository: Private - Company: DestinyX Studios ## Overview Forge is a local AI coding agent designed to provide a capable, private development workflow on Apple Silicon. The language model runs through Ollama, while Forge provides the orchestration layer around it: a full-screen terminal interface, route-aware agent loop, structured tools, safety approvals, repository retrieval, semantic memory, sessions, skills, MCP integrations, and optional LoRA specialization. Unlike a basic local-model chat interface, Forge can take controlled actions inside a development environment. It can inspect a repository, search code, edit files, run shell commands, execute tests, manage Git operations, create plans, delegate exploration to sub-agents, and recover file changes through checkpoints. ## Problem Cloud coding assistants are powerful, but they can introduce source-code privacy concerns, recurring API costs, provider dependency, limited control over memory, and opaque tool-execution behavior. The engineering question behind Forge was: > How can a capable coding agent run locally while still providing structured tools, repository understanding, persistent memory, context management, safety controls, extensibility, and a polished developer experience? ## Solution ### Terminal Interface A full-screen prompt_toolkit and Rich interface with a scrollable transcript, growing composer, live plans, tool status, model information, context usage, session controls, and mode switching. ### Routing and Product Layer Each turn is classified as social, local, research, or task. The route determines the prompt behavior, available tools, research workflow, and context strategy. ### Agent Runtime A multi-round ReAct loop prepares history, recalls relevant memories, calls the local model, executes approved tools, handles failures, and streams the final response. ### Tools and Safety Forge supports filesystem operations, shell commands, Git workflows, tests, patches, research, project retrieval, plans, sub-agents, security tools, skills, and MCP servers. Mutations are controlled by deterministic approval gates rather than relying only on model instructions. ### Local Models and Persistence Ollama runs local chat, coding, embedding, and vision models. SQLite stores sessions and episodes, while LanceDB stores semantic memory and project-scoped retrieval indexes. ## Capabilities ### Local Model Orchestration Runs open-weight coding, chat, embedding, and vision models locally through Ollama. ### Structured Tool Execution Allows models to read and edit files, run commands and tests, use Git, search the web, inspect images, update plans, and call MCP tools. ### Route-Aware Behavior Separates social conversation, local project analysis, research, and development tasks so each receives an appropriate prompt and tool allowlist. ### Human-in-the-Loop Safety Manual mode previews commands and changes before execution. Plan mode denies mutations. Security and elevated operations require stronger approval. ### Project RAG Indexes project files into LanceDB and retrieves relevant code based on semantic similarity. ### Persistent Memory Stores named sessions and conversation episodes in SQLite while keeping semantic facts in a separate vector memory layer. ### Context Engineering Prunes stale tool output, compacts long histories into structured summaries, preserves recent turns, and reserves context space for generation. ### MCP Extensibility Connects external Model Context Protocol servers and exposes their tools to the agent dynamically. ### Sub-agents Delegates repository exploration and security analysis to isolated agents without flooding the primary context. ### Optional Model Specialization Supports an MLX LoRA path for adapting compatible models to personal coding conventions on Apple Silicon. ## Design Decisions ### Use Ollama's native chat endpoint for tool turns. Some OpenAI-compatible Ollama endpoints may return Qwen tool calls as malformed XML or omit structured tool-call fields. Native Ollama chat provides more reliable tool behavior, while the OpenAI-compatible endpoint remains useful for non-tool requests. ### Enforce safety outside the model prompt. Manual, automatic, planning, and ask modes are enforced by application code. File mutations, shell commands, Git operations, sudo, and security tools have deterministic gates that cannot be bypassed through persuasive prompting alone. ### Separate sessions from episodes. A session represents a named long-term conversation bucket. Episodes represent isolated conversation segments so previous context does not silently bleed into every new launch. ### Separate semantic memory from project RAG. Personal facts and reusable preferences have different retrieval and lifecycle requirements from indexed source-code chunks. Forge stores them in separate LanceDB collections. ### Route turns before exposing tools. Social turns receive no tools, local questions receive read-oriented project tools, research receives web tools, and development tasks receive the full approved capability set. This reduces irrelevant tool calls and limits the model's action surface. ### Delay streamed prose during potential tool calls. Forge holds initial response deltas until it knows whether the model is making a tool call. This prevents the interface from printing a conversational preamble and then duplicating the real answer after tool execution. ### Use structured compaction instead of blind truncation. When context grows too large, Forge preserves intent, files, decisions, facts, and unresolved tasks instead of simply dropping the oldest messages. ### Isolate banter from coding history. Social turns use a separate no-tool prompt and a scrubbed history view so expressive responses do not contaminate later technical work. ## Safety Modes ### Manual Shows command or diff previews and requires approval before mutations. ### Auto Approves ordinary development mutations while retaining restrictions on elevated and security-sensitive actions. ### Plan Read-only. Mutating tools are denied regardless of what the model requests. ### Ask Pushes the agent toward clarification before taking action. ## Memory ### SQLite Sessions Stores session, timestamp, role, message content, and episode identifiers. ### Semantic Memory Uses nomic-embed-text and LanceDB to recall relevant long-term facts. ### Project RAG Indexes code and documentation in overlapping chunks scoped to the current project root. ### Automatic Indexing Projects are indexed automatically when they contain recognizable project markers. Very large repositories require an explicit indexing command. ### Context Compaction Preserves recent messages while converting older context into a structured summary when the token budget is exceeded. ## Public Release Vision Forge is being developed for public release as a local-first coding agent. Users bring an open-weight model that fits their hardware, while Forge provides the surrounding agent runtime: repository understanding, structured tools, reusable skills, memory, safety controls, and development workflows. Core inference runs locally through Ollama, so Forge does not require a cloud-model subscription for everyday use. Developers remain in control of their source code, model selection, data, and compute environment. Forge can inspect codebases, edit files, run commands and tests, manage Git workflows, retrieve project context through RAG, maintain session memory, and connect to external tools through MCP. Human approval gates remain available for file changes, commands, and other sensitive actions. The long-term direction is simple: a capable developer agent that runs on your machine, works with the model you choose, and remains free to use. ## Highlights - Python 3.13+ - Maximum 12 main-agent tool rounds - Maximum 8 read-only sub-agent rounds - Maximum 12 security-audit rounds - Four execution modes - Four route categories - SQLite session persistence - LanceDB semantic memory and project RAG - MCP tool integration - Local embedding and vision support - Optional MLX LoRA workflow - Tool-result limits and repeated-call detection - Automatic project indexing with large-repository safeguards ## Technology ### Runtime - Python 3.13 - Typer ### Interface - prompt_toolkit - Rich ### Models - Ollama - Qwen3 Coder - Qwen2.5-VL - nomic-embed-text ### Persistence - SQLite - LanceDB ### Agent Infrastructure - Structured tool calling - ReAct loop - MCP - Skills - Sub-agents ### Apple Silicon ML - MLX - mlx-lm - Optional LoRA adapters ### Developer Operations - Git - Docker sandboxing - gitleaks - Automated tests ## Status Forge is an active independent engineering project currently at version 0.1.0. The production repository remains private because it contains personal development workflows, security tooling, local configuration conventions, and experimental agent infrastructure. A public case study, architecture documentation, screenshots, and demonstrations are provided instead of unrestricted source-code access. ## Next Steps - Ship a 60–90 second demo: repository inspection → plan → file edit → approval gate → tests → summary - Build a repeatable agent evaluation suite - Measure tool-call success and task-completion rates - Compare workflows with and without project RAG - Improve broad repository exploration through skim_codebase - Add more granular checkpoint persistence - Expand MCP compatibility and diagnostics - Improve local-model fallback and capability detection - Benchmark cold-start and warmed-model latency - Package a free public CLI distribution of the agent runtime Author: Andres Echeverria (estebanech) --- # Saki > Saki is a private Discord assistant combining permission-aware LLM tools, deterministic moderation, SQLite memory, scheduled workflows, multimodal input, and Lavalink music. - HTML: https://estebanech.com/projects/saki - Markdown: https://estebanech.com/projects/saki.md ## Permission-Aware AI Agent for Discord Saki is a private, single-guild Discord assistant built as a persistent Gateway service. It combines a structured tool-calling LLM agent with deterministic moderation, persistent memory, scheduled reminders, server administration, image understanding, and music playback. ## Metadata - Role: Sole Designer and Engineer - Type: Independent Production Project - Status: Live - Runtime: Persistent Discord Gateway Service - Repository: Private ## Overview Saki is designed for one private Discord community rather than as a multi-tenant SaaS product. A guild allowlist rejects requests from every server except its configured home guild. Unlike a traditional command bot, Saki can interpret natural-language requests and take controlled actions through structured tools. Depending on the requesting user's authorization level, the agent can search server knowledge, summarize conversations, create reminders and events, manage music, perform moderation, or operate Discord resources. AI is invoked only during explicit interactions such as mentions, replies, /ask, and summarization flows. Fast moderation, member onboarding, reminders, and command handling remain deterministic application services. Discord acts as the user interface, while the bot process owns command handling, authorization, agent orchestration, moderation, persistence, music, and service integration. ## Problem Traditional Discord bots are reliable but limited to predefined commands. General-purpose AI chatbots understand natural language but should not receive unrestricted access to moderation tools or server infrastructure. The engineering question behind Saki was: > How can an LLM understand natural-language requests and take useful actions inside Discord without receiving uncontrolled administrative authority? ## Solution ### Discord Gateway Interface A persistent discord.js process receives messages, slash commands, button interactions, member events, voice-state updates, and raw gateway events required by Lavalink. ### Explicit AI Invocation The LLM runs only when a user directly mentions Saki, replies to Saki, invokes /ask, or uses another AI-specific flow. It does not process every server message. ### Permission-Aware Agent The agent receives only the tool schemas available to the requesting user. Public, moderator, and owner tools are exposed separately. ### Defense-in-Depth Execution Permissions are checked when tools are exposed to the model and checked again when a tool executes. ### Persistent Memory and Knowledge SQLite stores guild settings, reminders, learned server facts, user memories, and searchable conversation history. ### Deterministic Moderation Spam bursts and malicious links are handled outside the model for predictable latency, cost, and enforcement. ### Independent Music Service Music playback runs through a separate Lavalink service so chat, commands, memory, and moderation remain available if music is offline. ## Capabilities ### Structured Tool Agent Executes Discord actions through typed function tools instead of giving the model unrestricted API access. ### Permission-Aware Operations Filters tools according to public, moderator, and owner authorization levels. ### Persistent Memory Stores personal facts, guild knowledge, reminders, and searchable conversation history. ### Deterministic Moderation Handles malicious links and message-burst spam without calling an LLM. ### Scheduled Workflows Creates reminders, polls, and Discord scheduled events. ### Multimodal Input Uses a separate vision-model path for image understanding and comparison requests. ### Model Resilience Supports model fallback, per-model cooldowns, rate-limit handling, and oversized-request recovery. ### Music Infrastructure Uses Lavalink for playback, queue controls, loop modes, shuffle, volume, empty-channel cleanup, and source resolution. ### Skills and Knowledge Loads Markdown playbooks and static guild documentation when relevant. ### Graceful Degradation Keeps moderation and normal commands available if the AI provider or music service is temporarily unavailable. ## Design Decisions ### Run as a persistent Discord Gateway process. Message events, member events, voice state, moderation, and Lavalink integration require a long-running connection rather than an interaction-only webhook architecture. ### Invoke AI only during explicit interactions. This reduces cost, unnecessary message processing, context noise, accidental responses, and privacy concerns. ### Keep moderation deterministic. Spam and malicious-link enforcement need predictable behavior independent of model availability or interpretation. ### Filter tools before inference. A user cannot persuade the model to call a capability it was never given. ### Re-check permissions during execution. Tool visibility is not a replacement for deterministic server-side authorization. ### Fail closed on malformed arguments. Invalid model-generated JSON must not silently become empty or default arguments for destructive operations. ### Use SQLite for a single-guild deployment. SQLite with WAL mode provides sufficient durable persistence without unnecessary distributed database infrastructure. ### Separate Lavalink from the bot process. Music failures should not bring down commands, moderation, memory, reminders, or AI interactions. ### Bound agent rounds and context. Tool-loop and context limits reduce duplicate actions, runaway requests, and provider token-limit failures. ### Restrict allowed mentions. AI-generated responses must not be able to mass-ping roles or @everyone. ## Permissions ### Public Can use approved knowledge, reminders, polls, personal memory, music, channel discovery, social interactions, and self-service tools. ### Moderator Can use moderation, message cleanup, slowmode, channel and role operations, voice moderation, event management, and server-fact tools. ### Owner Can access invites, emoji management, webhooks, AutoMod operations, and higher-risk administrative capabilities. ## Personality Saki has a configurable personality layer designed to make a utility bot feel native to its community rather than like a generic corporate assistant. The product presentation is K-pop idol themed, inspired by Sakura Miyawaki of LE SSERAFIM, including profile and banner visuals in that aesthetic. Guild-wide personality modes include Normal, Sarcastic, Overkill, and Cute. The persona is implemented separately from authorization and tool execution. Changing the personality does not grant additional permissions or alter deterministic moderation rules. ## Privacy The source repository is private because it contains server-specific configuration, private knowledge structures, moderation behavior, internal prompts, Discord resource identifiers, infrastructure conventions, and security-sensitive integrations. The public case study focuses on architecture, authorization, persistence, agent orchestration, deployment, and non-sensitive demonstrations. ## Highlights - TypeScript ESM runtime - Node.js 22+ - discord.js v14 - Single allowed guild - Approximately 29 slash commands - Three authorization levels - Maximum five agent tool rounds - Three memory tiers - SQLite WAL mode - FTS5-backed search - 15-second reminder polling - Maximum two images per applicable vision request - Separate bot and Lavalink services - Optional Message Content and Guild Members intents - Model fallback and cooldown handling - Graceful shutdown for reminders, Lavalink, Discord, and SQLite ## Technology ### Runtime - TypeScript - Node.js 22 - tsx - pnpm ### Discord - discord.js v14 - Discord Gateway - Slash commands - Interactive components ### AI - OpenAI SDK - Groq - OpenAI-compatible model APIs - Tool calling - Vision models - Model routing ### Persistence - better-sqlite3 - SQLite WAL - FTS5 ### Music - Lavalink 4 - lavalink-client ### Infrastructure - Docker - Railway - Persistent volume - Private service networking ## Status Saki is deployed as a persistent private-server assistant. Its AI agent, deterministic moderation, memory, reminders, commands, onboarding, and music systems are separated into modules so individual services can degrade without disabling the entire bot. ## Next Steps - Destructive-action confirmation workflows - More detailed tool-execution audit logs - Protected user, role, and channel configuration - Expanded permission-boundary tests - Structured privacy and data-export controls - Automated SQLite backup validation - Agent evaluation tasks - Improved observability and tool-success metrics - Memory provenance and expiration controls - Additional reusable server workflow skills Author: Andres Echeverria (estebanech) --- # DestinyX Studios > DestinyX Studios is a software product and AI engineering company. Owned products: Oppli and Forge. Selective partner engineering for private platforms, agents, automation, and custom infrastructure. - HTML: https://estebanech.com/projects/destinyx - Markdown: https://estebanech.com/projects/destinyx.md ## Building our own products. Engineering private systems for select partners. DestinyX Studios builds customer-facing products of its own and engineers private software systems for select partners, with growing specialization in agent infrastructure, developer tools, and AI-enabled applications. ## Metadata - Role: Lead Software Engineer & Founder - Type: Software Product & AI Engineering Company - Status: Active - Business Lines: Products · Partner Engineering - Products: Oppli · Forge - Current Focus: Products, Developer Tools & Agent Systems - Long-Term Direction: Model Specialization & AI Research - Site: destinyxstudios.com - GitHub: github.com/DestinyX-Studios-LLC ## Overview DestinyX Studios is a software product and AI engineering company. We build our own customer-facing products and developer tools, while partnering selectively with organizations that need private platforms, intelligent agents, automation, or custom software infrastructure. Our product portfolio currently includes Oppli and Forge. Through selected partner engagements, DestinyX can also design and deliver internal operating systems, agentic assistants, platform integrations, workflow automation, and full-stack applications tailored to a specific organization. DestinyX provides the engineering and operating structure behind software intended for long-term ownership — whether that software is a DestinyX product or a private system built for a partner. The company’s goal is to build useful software first and allow real product and partner needs to guide deeper work in agents, tools, model specialization, and future model development. ## Problem Independent products often begin as experiments, but experiments do not provide the structure required for long-term ownership. The same gap appears when organizations need private systems that generic platforms cannot serve. Once software reaches real users — customers or internal teams — it needs clear technical direction, maintained infrastructure, controlled releases, documented systems, security boundaries, and someone accountable when production fails. DestinyX was created around one question: > How can ambitious products and private systems be developed with the continuity and engineering discipline of a dedicated product company? ## Solution ### Owned product development DestinyX gives each company product a durable home for architecture, implementation, infrastructure, releases, and continued development — starting with Oppli and Forge. ### Selective partner engineering For a limited number of partners, DestinyX designs and delivers private systems around real workflows, operational constraints, and long-term technical ownership — not generic agency retainers. ### Hands-on technical leadership Architecture, product direction, and implementation stay closely connected. Partner and product decisions are informed by direct work across application code, backend systems, data, AI workflows, deployment, and production operations. ### Long-term technical direction DestinyX builds with existing technologies today while steadily developing greater ownership of the agent, tool, evaluation, and model layers behind future products and partner systems. ## DestinyX Products ### Oppli **Career intelligence for better job-search decisions** A customer-facing career intelligence platform that helps users understand their market position, evaluate opportunities, identify missing skills, and prepare stronger applications. [Visit Oppli](https://www.oppli.app) ### Forge **A local AI coding agent and developer runtime** A developer tool planned for public release that combines local models with repository understanding, structured tool execution, project retrieval, persistent memory, safety approvals, and extensibility. [View Forge case study](https://estebanech.com/projects/forge) ## Partner Engineering DestinyX builds more than its own product portfolio. For selected partners, the studio designs and develops private software systems that cannot be adequately served by generic platforms or disconnected subscriptions. These engagements are built around the partner’s actual workflows, operational constraints, data, and long-term technical needs. Partner systems may include internal operating platforms, AI-assisted workflows, private agents, real-time community tools, integrations, automation, and full-stack applications. ### Engineering lifecycle - Product and workflow definition - System architecture - Interface and application development - Backend and database engineering - AI agent and tool integration - External APIs, webhooks, and synchronization - Authentication and permissions - Cloud infrastructure and deployment - Maintenance and continued iteration ## Partner capabilities ### Internal operating systems Staff-only platforms that combine business workflows, operational data, permissions, reporting, integrations, and platform controls into one system. ### Private AI agents Single-tenant agents that understand organization-specific context, operate through permission-gated tools, retain controlled memory, and assist with real workflows — including community assistants, internal support agents, operations copilots, research agents, knowledge assistants, moderation systems, and workflow agents. ### Private community agents Private, single-community agents for Discord and similar real-time platforms. These systems can combine natural-language assistance, structured administrative tools, moderation, memory, scheduled workflows, knowledge retrieval, and entertainment features under organization-specific permission rules. This capability is informed by prior independent engineering work on real-time, permission-aware Discord agent systems. ### Integrations and automation Systems connecting existing platforms through APIs, webhooks, synchronization workers, scheduled jobs, and custom business logic — across commerce platforms, Discord, internal databases, social platforms, CRM systems, reporting tools, and document intake. ### Full-stack product development Mobile, web, backend, data, and infrastructure engineering for partners that need a complete product rather than an isolated feature. ## Agentic Systems DestinyX develops agents that can do more than generate text. These systems can retrieve organization-specific knowledge, maintain controlled memory, call structured tools, operate across external services, follow permission boundaries, and complete multi-step workflows. Agent systems are designed around deterministic permissions, validation, observability, and human control rather than unrestricted model access. - Private Discord and community agents - Internal operations assistants - Customer-support copilots - Document and invoice intake - Research and reporting agents - Developer tools - Workflow automation - Persistent agents for games and interactive environments ## Long-Term Technical Direction DestinyX intends to progressively own more of the intelligence layer behind its products and partner systems. In the near term, this means developing stronger agent runtimes, structured tools, evaluation systems, local-model support, and model-independent infrastructure. As product needs and technical capacity grow, DestinyX plans to explore fine-tuning and specialization of open-weight models. Original domain-specific model development remains a longer-term objective, dependent on sufficient data, evaluation maturity, compute, research capacity, and a clear product advantage. Future products may include games and interactive environments where intelligent agents are part of the core experience. ## Engineering Principles ### Build for continued ownership Products and partner systems are designed around deployment, maintenance, failure handling, and future iteration, not merely the first successful demonstration. ### Keep technical leadership hands-on Architecture, product direction, and implementation stay closely connected. Partner engagements are intentionally limited so that remains true. ### Keep truth deterministic Permissions, business rules, calculations, and critical decisions remain in application code. Models assist with interpretation, retrieval, extraction, and execution within controlled boundaries. ### Move deeper into AI when evidence supports it Prompting, tools, retrieval, and existing models come first. Model specialization is pursued when evaluations demonstrate a meaningful product or partner benefit. ## Build With DestinyX Need a private internal platform, agentic assistant, workflow automation system, or full-stack product? DestinyX works with a limited number of partners on software that requires custom architecture and sustained technical ownership. - [Discuss a Partner Project](mailto:estebanech16@destinyxstudios.com?subject=DestinyX%20Partner%20Project) - View Engineering Capabilities ## Technology - Owned products - Partner engineering - Agent systems - Private platforms - Developer tools ## Status DestinyX Studios is active as a product company with a selective partner-engineering practice. Owned products currently include Oppli and Forge. Oppli represents customer-facing product work; Forge develops local AI, developer tooling, agent runtimes, and structured execution. Partner engineering is available for organizations that need private platforms, agents, automation, or custom infrastructure — with engagements limited so technical leadership stays involved in architecture and implementation. Current priorities include product reliability, public releases, production infrastructure, agent evaluation, clearer shared engineering standards, and selective partner work that fits the company’s technical focus. ## Next Steps - Continue developing and releasing Oppli - Prepare Forge for a future public release - Formalize reusable agent, tool, and evaluation infrastructure - Expand local and open-weight model support - Begin measurable model-specialization experiments - Take on selective partner engagements under DestinyX - Explore future customer applications and games Author: Andres Echeverria (estebanech) --- # VidVortex > VidVortex is a cross-platform desktop media downloader built with Python, yt-dlp, and ffmpeg — with per-OS setup scripts and GitHub Releases packaging. - HTML: https://estebanech.com/projects/vidvortex - Markdown: https://estebanech.com/projects/vidvortex.md ## A Cross-Platform Media Downloader Friends Can Actually Run VidVortex is a desktop app for pulling audio or video from a URL — paste a link, choose video or audio, load qualities, and download to organized folders — without requiring users to memorize yt-dlp flags. ## Metadata - Role: Sole Designer and Engineer - Type: Open Source Tool - Status: Shipped - Distribution: GitHub Releases - Platforms: Windows · macOS · Linux - Repository: Public ## Overview VidVortex wraps the power of yt-dlp and ffmpeg in a small native-feeling desktop workflow. Users paste a link, choose video or audio, load available qualities, and download into organized output folders. The hard part was not the download call itself — it was making the toolchain reliable on real machines: first-run setup, PATH hygiene, frozen builds per OS, and clear Activity feedback when hosts (especially YouTube) need cookies or extra auth steps. ## Problem CLI media tools are flexible, but friends and family rarely want to install Python packages, resolve ffmpeg PATH issues, or decode format selectors before a simple download works. The engineering question behind VidVortex was: > How do you ship a yt-dlp/ffmpeg workflow as something people can run from GitHub Releases — with quality picking, setup UX, and supportable error feedback — without reading a man page first? ## Solution ### Desktop download flow A focused UI for URL input, media type selection, quality discovery, and download orchestration with in-app Activity/status feedback. ### Per-OS setup scripts First-run scripts install or refresh yt-dlp and ffmpeg on Windows, macOS (arm64/Intel), and Linux so the GUI inherits a working toolchain. ### Release packaging PyInstaller-style frozen builds and GitHub Actions produce versioned zips for each platform so distribution does not depend on users cloning the repo. ## Capabilities ### Quality-aware downloads Discover formats via yt-dlp, let the user pick quality, and route mux/remux work through ffmpeg instead of exposing raw CLI selectors. ### Host edge-case handling In-app guidance for tricky hosts — browser cookie hints, optional Netscape cookies.txt, and Activity logging for supportability. ### Cross-platform packaging Windows .exe, macOS launcher + .bin, and Linux binary layouts documented for GitHub Releases consumers. ### Toolchain hygiene Launcher scripts keep GUI runs on the same PATH users verify in Terminal — especially important on macOS security and Homebrew setups. ## Design Decisions ### Build on yt-dlp + ffmpeg instead of reimplementing extractors Host support and format discovery are moving targets. Wrapping battle-tested tools beats maintaining a custom extraction layer. ### Invest in first-run setup UX Most download failures for non-technical users are environment failures. Setup scripts and PATH hygiene matter more than extra UI chrome. ### Ship via GitHub Actions Releases Versioned platform zips let friends install without cloning, and CI keeps packaging reproducible across Windows, macOS, and Linux. ## Highlights - Python desktop UI + yt-dlp/ffmpeg orchestration - Per-OS setup scripts for dependency refresh - Quality picking without CLI flag gymnastics - Cookie/auth flows for YouTube-class hosts - PyInstaller-style frozen builds per platform - GitHub Actions release packaging ## Technology ### Languages & runtime - Python 3 — UI, orchestration, subprocess integration - Shell / Batch / PowerShell — setup and release automation ### Core stack - yt-dlp — extraction & format discovery - ffmpeg — mux/remux pipelines ### Packaging & delivery - PyInstaller-style frozen builds - GitHub Actions cross-platform releases - Windows · macOS arm64/Intel · Linux ## Status VidVortex is shipped as a public open-source tool. Builds are distributed through GitHub Releases for Windows, macOS, and Linux. Ongoing work focuses on keeping yt-dlp/ffmpeg paths reliable as host sites and packaging environments change. ## Next Steps - Keep release packaging current with yt-dlp and ffmpeg changes - Improve first-run diagnostics when toolchain detection fails - Refine Activity logging for faster support on auth/cookie edge cases Author: Andres Echeverria (estebanech)