I build tools that help AI agents remember, coordinate, and ship.
Local-first software · Agent infrastructure · Creative tools · Hardware
I’m Avidlive, building practical tools around a recurring problem: useful agent work should survive the next session, the next model, and the next tool.
My projects cover shared memory, coding-agent workspaces, model routing, verified execution loops, image workflows, and hardware development. I share builds, experiments, and guides on X · @Av1dlive.
| Project | What you can do with it |
|---|---|
| Agentic Stack | Carry one .agent/ memory-and-skills layer across coding tools. Run bounded work loops, inspect activity, and keep reusable lessons. |
| Agentic Stack Desktop | Search conversations across Claude Code, Codex, OpenCode, and Cursor; explore a local knowledge graph; work with custom agents, tasks, and terminals in a native macOS app. |
| Brain | Give agents shared, Git-backed memory through a CLI, terminal dashboard, and MCP server. Search notes and sync when you choose. |
| Continual Memory Policy Model | Explore how an agent chooses what to remember: six memory operations, delayed outcome feedback, a read-only Memory Center, and evaluated adapter training with human promotion. Experimental research. |
| Image Loop + Image Reconstruction | Break references into named parts, preserve chosen details, compile prompts, review edits, and repair failed requirements. Includes visual maps, reusable JSON, and worked examples. |
| Agent Harness | Give a coding agent a repeatable spec → plan → implement → verify → release workflow for SaaS, CLI, MCP, and mobile work. |
| GateFlow | Work on SystemVerilog, Verilog, and VHDL with agent-assisted design, linting, testbenches, simulation, formal checks, and FPGA tooling. |
| ztk | Reduce the command-output noise entering an agent’s context, integrate with supported agent hooks, and inspect savings locally. |
| Meridian OS | Explore a company console for goals, tasks, agents, budgets, approvals, and activity. Includes a simulation mode and optional runtime connections. |
- Memory and knowledge: ContextHub, Familiar Second Brain, Design Brain, and Graph Engineering.
- Model routing and execution: SageRoute, Quota Flow, Fable Orchestrator, and Scoped Agent Loop.
- Writing and creative work: X Article Skills, Prompt Skills, and RosterRoom.
- Everyday tools: LocalFlow for on-device dictation, Cove for a local productivity workspace, and spotify-kitty for music in the terminal.
Keep context portable. Make decisions inspectable. Verify the result.
That means local data where practical, reusable files and open interfaces, bounded execution, and human control over consequential actions. Some repositories are working tools; others are prototypes or research experiments. Each project’s README explains its setup, evidence, and limits.
Browse the full public project directory
- agentic-stack — Portable memory, skills, protocols, and bounded loops across coding-agent tools.
- agentic-stack-desktop — Native macOS workspace with shared conversation search, a knowledge graph, custom agents, and terminals.
- brain — Git-backed agent memory with a CLI, terminal dashboard, MCP server, and explicit sync.
- continual-memory-policy-model — Experimental memory policies with delayed feedback, a local Memory Center, and gated adapter training.
- contexthub — Local MCP context layer with connectors, line-level retrieval, and scope filtering.
- familiar-second-brain — Markdown-vault memory, Kimi session capture, daily briefs, and MCP access.
- design-brain — Capture design tokens and patterns from websites and screenshots; search, compare, and export them.
- agent-harness — File-based workflows for taking SaaS, CLI, MCP, and mobile projects from specification to verification.
- agentic-harness — Rust-native agent runtime, SDK, and CLI.
- autoharness — Local macOS control center for direct tasks, bounded loops, swarms, and task graphs.
- sageroute — OpenAI-compatible model routing that escalates based on agent execution evidence.
- quota-flow — Codex orchestration skill with implementation, focused review, and verified repair loops.
- fable-orchestrator — Fable-led orchestration with focused implementation workers.
- codex-self-improving-loop — Task execution, grading, and prompt revision in a repeatable Codex loop.
- scoped-agent-loop — Two-agent planner/executor loop with explicit policy and execution receipts.
- claude-model-switch — Provider switching, custom API keys, and multi-model tmux orchestration for Claude Code.
- dsh-rs — Rust/GPUI desktop harness project.
- ztk — Compact command output before it enters an agent’s context; inspect compression statistics.
- tokensavings-pro — Reusable dual-model workflow with compact implementation and reasoning packets.
- image-loop — Reference-driven image creation and editing with visual maps, review, and bounded repairs.
- image-reconstruction — Map a reference into named parts, reusable JSON, and reconstruction or edit prompts.
- x-article-skills — Skills for researching, writing, and packaging articles on X.
- prompt-skills — Reusable prompt templates for writing, analysis, technical work, and productivity.
- prompting-skill — Portable workflow for designing and improving LLM prompts.
- avids-essential-skills — A collection of practical Claude Code skills.
- rosterroom — Ready-to-use team rosters for Grok Bot.
- ste100-loop — Draft, grade, and revise technical writing with a clarity-focused review loop.
- meridian-company-os — Company console with goals, tasks, agents, budgets, approvals, and a built-in simulation.
- company-foundry — Company-harness methods, agent presets, and operating artifacts.
- agentic-os — Engineering-activity knowledge graph and execution-drift detection MVP.
- business-workflow-autopilot — Agent skill for turning business operations into structured workflows.
- graph-engineering — Knowledge-graph pipelines, task-graph patterns, teaching mode, and reusable workflows.
- agent-roadmap-2026 — A staged learning roadmap for AI agent engineering, grounded in primary sources.
- larp-bench — Research benchmark exploring models that invent personal experiences.
- murmur — Multi-agent social-media simulation engine.
- voice-agent-builder — Builder’s guide and executable skill for voice-agent development.
- Gateflow-Plugin — Claude Code hardware-development plugin with HDL design, simulation, formal verification, and FPGA workflows.
- gateflow-cli — AI-assisted SystemVerilog development environment.
- LocalFlow — On-device dictation for Apple silicon Macs.
- cove — Local browser workspace for tasks, notes, pages, habits, and journaling.
- spotify-kitty — Spotify terminal client with inline album art, search, playlists, and playback controls.
- homebrew-tap — Homebrew distribution for my tools.
- homebrew-ztk — Homebrew tap for ztk.
- homebrew-contexthub — Homebrew tap for ContextHub.
Public forks and upstream projects
These are forks of other projects. Original authorship and licenses belong to their respective upstream projects.
Follow the builds: X / @Av1dlive · All repositories
Found a bug or have an idea? Open an issue in the relevant project.