Sven

Sends messages between agents, broadcasts to channels, and retrieves message history for inter-agent communication.

Spawns, manages lifecycle, and terminates autonomous AI agents from the 40+ built-in agent catalog.

Optimize prompts and context windows to reduce token usage while preserving quality.

Animated desktop character state machine for Sven's Tauri companion app. Manages character form (ORB, ARIA, REX, ORION), state transitions (idle→thinking→speaking→celebrating), walk cycles, thought bubbles, sound effects, and real-time agent event sync.

Multi-model deliberation system. Sends queries to multiple LLMs simultaneously, has them peer-review each other's responses anonymously, then a chairman model synthesizes the best answer. Supports configurable council composition, voting strategies, and cost tracking.

Interactive educational autograd engine — port of Karpathy's micrograd. Build, train, and visualise tiny neural networks step-by-step to learn how backpropagation works.

Runs benchmark suites against models, tracks ELO rankings, manages A/B tests, and generates performance reports.

Manages the model catalog — register, unregister, query models by task or status, set defaults, and check health.

Routes inference requests to the optimal available model based on task type, quality priority, latency budget, and VRAM constraints.

Fine-tuning pipeline for small open models using LoRA/QLoRA. Trains domain-specific adapters from conversation logs, codebase files, or custom datasets. Pre-built recipes for writing style, codebase conventions, and domain vocabulary. Exports to LiteLLM for immediate use.

Compress long prompts and contexts while preserving semantic meaning — reduce token count without losing information.

Teaches Sven how to extend himself — create new skills, write handlers, understand codebase conventions, and use the skill authoring system. This is Sven's guide to coding himself.

Teaches Sven how to use his own self-healing pipeline — diagnose issues, scan code, propose fixes, deploy, rollback, and introspect heal history. This is Sven's operational playbook for maintaining himself.

Introspects Sven's own architecture, capabilities, codebase structure, service topology, and current operational state. Use when Sven needs to describe himself, explain what he can do, or reference his own systems.

AI-powered driver assistant and legal shield for 47Plate — diagnoses warning lights from photos, generates ticket appeal PDFs with dashcam evidence, checks STAS 1848 road sign compliance, reads OBD-II diagnostic trouble codes, and OCR-extracts receipts/invoices for TCO dashboards.

Analyze SQL queries — suggest indexes, detect N+1 patterns, explain execution plans.

Analyze and optimize Dockerfiles — reduce image size, improve layer caching, fix security issues.

Plans AirLLM-style layer-by-layer inference across mesh devices for running large models on limited hardware.

Manages the compute mesh device registry — register, query, heartbeat, opt-in/out, and view aggregate mesh stats.

Creates, monitors, and manages distributed compute jobs with MapReduce, Pipeline, ScatterGather, and LayerSplit strategies.

Schedules work units to mesh devices based on capability matching, load balancing, battery awareness, and priority.

Generate color palettes, evaluate contrast ratios, simulate color blindness, and create full light/dark themes from a single seed color. Uses OKLCH for perceptual uniformity.

Audit designs for quality issues — contrast, spacing, typography, accessibility, color blindness safety. Provides actionable improvement suggestions with severity scoring.

Generate spacing scales, CSS grid systems, responsive breakpoints, z-index layers, and ready-to-use layout patterns (dashboard, sidebar, card grid, holy grail, etc).

Generate CSS animations with proper easing, timing, spring physics, stagger patterns, and prefers-reduced-motion accessibility fallbacks.

Generate modular type scales, recommend font pairings by mood, create fluid responsive typography, and analyze text readability metrics.

Create videos programmatically — marketing videos, data visualizations, social media content, tutorials, and brand videos using ffmpeg-based rendering with declarative JSON composition specs.

Define and manage scheduled email workflows — drip campaigns, follow-up sequences, and recurring email tasks.

Draft professional emails from context — compose, format, suggest subject lines and tone.

Generate contextual email replies — analyze incoming email and produce an appropriate response.

Send/search emails via a generic IMAP/SMTP bridge (or vendor API).

Generate images via OpenAI (DALL-E / gpt-image-1) or Stable Diffusion (AUTOMATIC1111-compatible) endpoints.

Manage Sven's proactive notification configuration — enable/disable notifications, set quiet hours, configure cooldowns, manage trigger rules and delivery channels.

Send proactive messages to the user through configured channels. Allows Sven to autonomously reach out with questions, alerts, progress updates, or any message without waiting for user input.

Schedule a message to be sent to the user at a specific future time via the Sven companion app. Use this when the user asks Sven to send them a reminder, notification, or any message at a specific time.

Processes multiple documents through the OCR pipeline in batch with progress tracking.

Extracts code from screenshots using OCR with code-aware detection and syntax preservation.

Extracts text from documents and images using GLM-OCR with multi-language, table, handwriting, math, and code detection.

Searches across processed documents for keywords, entities, and semantic matches.

Generates extractive summaries from documents with configurable style, key point extraction, and document comparison.