Forkcast Self-Hosted Projects #1: mempalace, hermes-webui, reclip, agentic-inbox, ministack

Episode Summary#

The debut self-hosted episode explores five projects you can run on your own infrastructure. From the highest-scoring AI memory system ever benchmarked to a browser-based interface for Hermes Agent, a clipboard management system, an email-based agent interface, and a minimal full-stack deployment template — all designed for privacy-conscious self-hosted environments.

Repositories Covered#

mempalace#

The highest-scoring AI memory system ever benchmarked (96.6% R@5 on LongMemEval). Implements a hierarchical memory palace structure with 30x AAAK compression, Hebbian potentiation, and Ebbinghaus decay. Runs locally with ChromaDB and SQLite — no cloud dependencies.

hermes-webui#

A browser-based chat interface for Nous Research’s Hermes Agent. Manage sessions, view usage analytics, configure multi-platform integrations (Telegram, Discord, Slack, WhatsApp), manage cron jobs, and browse agent files — all through a clean web dashboard.

G0DM0D3#

Most AI chat interfaces have heavy-handed content filtering and moderation that limits what you can discuss Self-hosted, open-source, uncensored chat interface

reclip#

A self-hosted clipboard management and synchronization tool. Captures, organizes, and syncs clipboard history across devices with full-text search, tagging, and secure encrypted storage — all running on your own server.

agentic-inbox#

An email-based interface for AI agents that lets you interact with autonomous assistants through your existing email workflow. Agents process incoming messages, execute tasks, and respond — turning your inbox into a command center for agentic automation.

ministack#

A minimal, opinionated full-stack deployment template for self-hosted applications. Combines Docker Compose, Traefik for reverse proxying with automatic SSL, and a curated set of services that give you production-grade infrastructure in a single repository.

claude-code-local#

Using Claude Code means sending your codebase to the cloud — a non-starter for NDA-protected, legal, or healthcare workflows 65 tok/s on Qwen 3.5 122B, supports Llama 3.3 70B and Gemma 4 31B, MLX-native, airgap-ready, Apple Silicon

HolyClaude#

Switching between Claude Code, web UIs, headless browsers, and CLI tools fragments your AI workflow across six different windows Claude Code + web UI + 7 AI CLIs + headless browser, 50+ integrated tools, Docker-based

cabinet#

Your startup’s knowledge is scattered across Notion, Slack, email, and a dozen SaaS tools — there’s no single source of truth AI-first knowledge management, startup OS, self-hosted

brightbean-studio#

Social media management tools like Buffer and Sendible charge $30 to $100 a month for basic scheduling across platforms 10+ platforms, single dashboard, free alternative to Buffer/Sendible/SocialPilot, self-hosted

leaf#

Previewing Markdown in the terminal usually looks like raw text — you lose all sense of structure until you render it in a browser GUI-like Markdown preview in terminal, self-hosted

future-agi#

Evaluating LLM applications in production is a mess — tracing, evals, datasets, and guardrails live in five different tools Tracing, evals, simulations, datasets, gateway, guardrails — all in one platform, Apache 2.0

arkon#

Every employee connecting their AI coding tool to your infrastructure creates a governance nightmare — who has access to what, and which models are they using Centralized MCP access control, resource management, workspace contexts, single portal for admins

hermes-control-interface#

Running AI agents on a server is powerful, but managing them through SSH alone is painful — no visibility into sessions, cron jobs, or system health Browser-based terminal, file explorer, session overview, cron management, system metrics, agent status panel, single password gate

obsidian-llm-wiki-local#

You want a personal AI wiki but sending your private notes to a cloud API defeats the purpose of a personal knowledge base 100% local with Ollama, auto-extracts concepts from Markdown, Obsidian integration, zero data sharing

tome-app#

Meeting transcription tools send your confidential discussions to the cloud — a dealbreaker for legal, medical, or strategic meetings Local transcription, exports to Obsidian vault, no cloud, no API keys required

openclaw-managed-agents#

Claude Managed Agents and ChatGPT Workspace Agents lock you into specific models and cloud providers Any model, any cloud, open source alternative to Claude Managed Agents and ChatGPT Workspace Agents

graymatter#

Your AI agents amnesia between conversations means you burn tokens repeating context, sometimes wasting 90% of your API budget on redundant prompts 3 lines of code to integrate, reduces token consumption by 90%, persistent agent memory

signal#

Managing AI agents across different providers and tools creates fragmentation — you need a unified interface Self-hosted, multi-provider AI agent management, unified interface

OpenPostings#

Job hunting means maintaining profiles and checking postings across Greenhouse, Lever, Workday, and a dozen other ATS platforms — each with its own login and UI Open source ATS aggregation, self-hosted, MCP support

scrumboy#

Project management tools like Jira and Linear are cloud-only, expensive at scale, and overkill for small teams that just want a kanban board Shareable boards, voice commands, sticky notes, MCP support, self-hosted

zano#

AI agents work in isolation — you run one in your terminal, another in the browser, and neither knows what the other is doing Shared channels for humans and AI agents, collaborative workspace, self-hosted

saasmail#

Setting up transactional email for a SaaS product means either paying SendGrid’s markup or wrestling with SMTP server config for a week Cloudflare Workers-based, self-hosted, built for SaaS teams

flowboard#

Creating AI product demo videos requires bouncing between storyboarding tools, video editors, and AI generators — a slow, fragmented process Infinite canvas, drag-and-drop nodes, auto-prompt generation, Google Flow + Claude CLI, runs locally

SkySend#

File sharing services like WeTransfer upload your files to their servers unencrypted — fine for cat photos, terrible for sensitive documents End-to-end encrypted, zero-knowledge server, browser-side encryption, no accounts, no telemetry

pullmd#

Converting web pages to clean Markdown for AI agents usually means either fragile scraping scripts or paying for a SaaS converter that sees all your URLs PWA + REST + MCP + Claude Code skill, Reddit support, refreshable share links, self-hosted

HA-Optimizer#

Home Assistant optimizer for smart home automation configurations

karadul#

Setting up a VPN between your home lab, VPS, and office usually means WireGuard config files, dynamic DNS, and a weekend of debugging MTU issues Zero dependencies, mesh VPN, written in Go, single binary

knowledge-engine#

The bridge between human-readable wikis and machine-speed memory. Built on Karpathy’s LLM Wiki pattern + Memvid.

4DPocket#

You save content across Reddit, Twitter, YouTube, and a dozen other platforms — but finding anything later means searching each one individually 17+ platforms, auto-tagging, AI summarization, semantic search, self-hosted

api#

Self-hosted image background removal API for AI-powered photo editing

posta#

Sending email from your app shouldn’t require a third-party API that reads your users’ emails and charges per message HTTP API + SMTP delivery, templates, analytics, self-hosted alternative to SendGrid/Mailgun

sverklo#

AI coding assistants waste tokens and context window on irrelevant files because they can’t understand your codebase structure 43× fewer tokens than grep, symbol graph, blast-radius analysis, git-pinned memory, zero config, MIT licensed

Watch#

• Video: https://www.youtube.com/watch?v=-nwjtYFcbgo
• Cover: https://i.ytimg.com/vi/-nwjtYFcbgo/hqdefault.jpg

Notes#

Transcript and notes will be added from Forkcast output artifacts.