CLIProxyAPI Dashboard puts a web console in front of CLIProxyAPI so you manage AI providers, keys, quotas, and containers through forms instead of YAML files. Connect Claude Code, Codex, xAI, Antigravity, Kimi, and Devin accounts with OAuth, then add Gemini CLI, GitHub Copilot, Kiro, and Cursor through plugins you install and configure from the settings store. Any OpenAI-compatible endpoint can join as a custom provider with model mappings, and admins can mark a provider as shared so the whole team uses it. You create, revoke, and track API keys with per-user ownership, manage dashboard users and admin rights, and stream live logs beside container health. Quota views track rate limits per provider and per model, and Telegram alerts fire when remaining quota drops below your threshold. Usage analytics break down request counts, provider share, model statistics, error rates, and estimated cost. Configuration syncs to OpenCode automatically, model configurations travel as share codes, and backups can be created, scheduled, previewed, and restored from settings. One-click updates refresh the dashboard and proxy images, containers start and stop from the panel, and automatic TLS keeps the console reachable under your domain. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Your Telegram chats already contain your tasks, agreements, and reminders; Iva Agent turns that stream into a memory you can query months later from the same chat window. Send a voice note and it returns a task list, a draft email, or a meeting card, with the original words stored verbatim in a dated markdown log. A nightly rollup folds each day into weekly, monthly, and yearly summaries, then rewrites changing facts into typed cards for contacts, projects, decisions, and ideas. Ask what you agreed with a client and it answers from the card it wrote, complete with date and context. Photos are described by your vision model, forwarded posts pass an injection screen before the model reads them, and outgoing messages pass a secret-redaction gate. You can connect Gmail, Calendar, Drive, Sheets, Docs, and Tasks, search the web through Tavily, Exa, Parallel, or Brave, and add skills or MCP servers by dropping in a file. An optional userbot reads group chats from your own account, folds them into summaries, and flags messages that need you, guarded by server-enforced send pacing. Token accounting logs every model step, and the user allowlist fails closed when empty. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

1FlowBase gives teams a single self-hosted foundation that connects model providers, retains every AI conversation, and turns that activity into working business applications. You connect OpenAI, Anthropic, Gemini, DeepSeek, Alibaba Cloud Bailian, or any OpenAI-compatible API through provider plugins and publish one stable endpoint for clients. Requests convert between Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses, so applications keep working when a provider changes. You can route traffic by model name, fall back across branches when a provider fails, and combine planning, execution, and review models into a virtual model that callers use like any other model. Every gateway conversation is recorded with its model, reasoning effort, status, token totals, recorded cost, and latency, then broken down by model and by user in statistics views. You define data models and data sources, expose CRUD APIs, schedule workflows, and publish an OpenAPI 3.1 document for each application. Agents reach the same runtime through MCP with progressive list, get, and call tool discovery, and you manage MCP instances and third-party servers from the console. Pages come from native React blocks that you edit in place with a built-in AI assistant. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Bloom hires you a private tutor for any subject, then teaches one lesson at a time and rewrites the next lesson around what you actually understood. You begin with a topic, an uploaded PDF, text, or Markdown source, or a set of project files, and Bloom drafts a syllabus of concrete mastery items at simple, standard, or deep expansion. Each lesson arrives focused and readable, with a chapter sidebar for quick jumps and a collapsible syllabus tracking what is already mastered. Reading is active work: highlight any passage and a question thread opens beside it, streaming an answer that stays attached to that sentence for later follow-ups. Reflection questions close every lesson, and your answers, annotations, and highlighted gaps become the raw material for the next lesson, which opens by reviewing the previous answers. Source mode keeps page numbers and sections so lessons can cite where an idea came from, and project files render each document as its own page for highlight questions without generated lessons. Finished courses produce an evaluation and a summary, followed by three recommended next topics drawn from your learning history. A personal center shows a study calendar, daily activity, and streaks across your courses. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Mix Studio gives ComfyUI a calm, touch-friendly workspace where image and video generation happens through focused controls instead of a fresh node graph for every job. The Create view runs curated families including Krea 2, Flux 2 Klein, and Qwen Image with exact resolution, seed, batching, LoRA, and prompt enhancement controls. Region mode places aspect-correct boxes on the canvas, each carrying an independent prompt, LoRA stack, and reference image, then resolves them into one coherent generation. Edit mode combines whole-image editing, identity editing, and multi-reference Remix flows with masks, inpainting, outpainting, and source preservation. Video workspaces cover text-to-video, first and last frame control, reference-to-video, character animation, face lipsync, and motion transfer driven by a trimmed performance clip. An experimental photo-to-3D path turns a picture into an orbit video and a Gaussian splat you can view interactively and export. Upscaling pairs synchronized pan and zoom with a movable before-and-after divider for detail inspection. The Library keeps generations and uploads searchable with folders, groups, metadata, reusable settings, recoverable trash, and ZIP export, and separate profiles can carry PIN access. Live queue progress, cancellation, guided setup with dependency checks, and automatic backups keep daily operation orderly. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. GPL-3.0 licensed.

Immich Upload Optimizer shrinks every photo and video at the moment of upload, so a self-hosted Immich library stops growing faster than the disk beneath it. The proxy sits in front of your Immich server and receives uploads first, matching each file by extension against a task list you define in a YAML file. Images pass through JPEG-XL or Caesium command line tools for lossless optimization by default, preserving every pixel of the original capture. Owners who prefer smaller files can switch a task to lossy settings that cut image size dramatically at the same perceived quality. Video conversion ships disabled because lossless transcoding rarely wins space, and owners can enable HandBrake profiles for lossy conversion with strong savings. Tasks run in sequence with fallback: the first matching command that succeeds supplies the stored file, and an upload with no matching task is refused rather than stored unprocessed. Long conversions keep working in the background after a client disconnects, and the processed file still reaches Immich. Deployment adds one service to an existing Docker Compose stack, points clients at the proxy port, and removes the direct public port from the Immich server itself. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

ProductFlow turns a product photo and a few facts into finished selling material, giving solo merchants and small teams one workspace for copy, posters, and repeatable image generation. You start by creating a product record, attaching original and reference images, then arranging product data, copy nodes, and image nodes on a zoomable canvas with drag, pan, multi-select, and connections. Copy nodes generate editable structured text that you can review, confirm, and keep in history before downstream image nodes read it. Poster generation runs in template mode with local rendering or generated mode through an image provider, and completed results can be downloaded, saved as product references, or promoted to the main product image slot. A separate image session view supports text-to-image and image-to-image runs with base image selection, candidate comparison, queue position, progress, cancel, and retry. Every finished image can move into the gallery with its source session, linked product, prompt, size, and model recorded for later reuse. Runtime settings cover provider profiles, model bindings, image tool parameters, prompt templates, upload limits, and concurrency, guarded by a separate settings token. The built-in help pages document the canvas, gallery, image sessions, and configuration in searchable sections. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Guaardvark is a complete AI studio that lives on your own GPU, so chat, retrieval, agents and media generation happen on hardware you own instead of a metered cloud account. Upload documents and question them through local retrieval that fuses keyword and vector search, reranks candidates, and cites the passages behind each answer, with the chunks and scores visible for inspection. Plan mode turns a request into an approve-or-edit plan card before tools run, and approved steps execute one at a time with a pause on failure. Screen agents receive a desktop of their own, see it through a vision model, and operate a real browser while the operator watches from any page. Coding swarms split a task across parallel agents in separate git worktrees and merge results back in dependency order. The generator bench covers thirteen local video models, image models for generation, editing, inpainting and upscaling, full songs with vocals, sound effects and three text-to-speech engines, with consent required before any voice clone or public post. Character, environment and prop LoRAs train from a few reference photos and keep the same character consistent across images and films. A GPU orchestrator with a live VRAM budget schedules renders, chat and training so jobs stop competing for the card. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Expose a service running on your laptop to the public internet in one command, through a tunnel server that you control. mmar pairs a small client with a self-hosted server: run the client against a local port and it registers a tunnel, receives a generated subdomain, and prints the public URL that now forwards to your machine. Every request that arrives shows up in live client logs, so a webhook test or a shared preview carries a visible trail of hits as they happen. Multiple local ports can be exposed on different subdomains at the same time, which suits a frontend, an API and a callback receiver running side by side during development. On the server side, two listeners do the work, one for HTTP traffic and one for the TCP tunnel connections, and a protected stats page returns JSON listing connected clients, their subdomains and creation times. Wildcard DNS plus a Caddy layer4 proxy with automatic wildcard certificates turns a personal domain into tunnel territory, and credential hashes guard the stats endpoint. The hosted mmar.dev service offers a free starting point on a generated subdomain, and the same single binary moves the whole arrangement onto your own infrastructure when control matters more. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

Give every AI agent you use the same memory, and a coding session stops starting from zero each time the tool changes. Signet runs as a daemon that captures transcripts, imported documents and connected sources, then builds a structured map of the people, projects, facts and relationships inside that history through a background process called dreaming. Each connection links back to its source passage, so an agent can trace a recalled fact to the exact transcript or file that produced it. Hooks, plugins and MCP integrations feed relevant memory into Claude Code, Codex, OpenCode, Gemini CLI, Kimi Code, Pi and more before the next prompt begins, and the same memory follows the work when the harness changes. Agents can also leave messages for each other through Signet, delivered at the start of the recipient's next session. The browser dashboard presents a daily brief, recent memories, a searchable memory graph, source connections and system status, plus a chat that answers questions over the memory with citations to the underlying records. Imports cover Word, PowerPoint, Excel, PDF, EPUB, CSV and web pages, with Notion databases and Discord channels syncing into the same graph. Per-agent read isolation, encrypted secrets and credential redaction keep sensitive material bounded. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

CatGPT Gateway turns the ChatGPT or Claude account a team already pays for into an OpenAI-compatible API that ordinary tools can call. Point the OpenAI Python SDK, LangChain, or a plain curl request at the gateway and choose the browser-backed model for Claude or ChatGPT; the service opens the provider web interface in a real Chromium session, enters the prompt, waits for completion, extracts the answer, and returns standard OpenAI JSON. Streaming responses arrive over SSE, persistent session headers keep multi-turn conversations on one browser tab to avoid reloads, and a browser page pool handles concurrent requests across several tabs. Tool calling round-trips through structured prompt injection, so LangChain agents can invoke local functions and receive parsed tool calls in the normal format. Vision inputs, PDF and document attachments, and ChatGPT image generation through DALL-E pass through the same endpoint shape that existing clients already speak. A live monitor at the preview route shows every active browser tab in a double-buffered grid, with tab switching and safe tab closure for the operator. One browser login through a noVNC page persists in a Docker volume, so the API keeps serving after restarts without storing provider API keys. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Containarium gives every AI agent a persistent Linux box of its own, reachable over SSH and driven through structured tools instead of a fragile terminal session. One command creates a named sandbox, syncs the SSH configuration, and wires the box into Claude Code, Codex, Gemini or Cursor through an MCP server that runs inside the container. From there the agent can execute shell commands, read and write files, install services, and expose a port on a public hostname with TLS terminated by a Caddy sentinel. Detached code runs keep working after a laptop sleeps or a connection drops, and the operator reattaches to the same byte-exact log later, checks status, or stops the run without losing output. An optional per-tenant eBPF egress policy limits outbound traffic per box, logs denied attempts for audit, and keeps provider API keys inside a model gateway rather than inside the agent environment. Boxes run on LXC or Kubernetes, keep state between agent sessions, support GPU passthrough for local model work, and stay separate from the host operating system and the cluster control plane. The dashboard presents containers, traffic, security events, audit records, GPU state and monitoring in one place. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

AI Real Estate Assistant puts a conversational search engine on top of your own property catalogue, so buyers type a request in plain language and receive matched listings in return. A query about a two-bedroom apartment in a named city under a price limit becomes extracted filters, semantic ranking over the catalogue, and a shortlist that matches the words instead of a rigid form. Behind the chat sits a working toolbelt: a mortgage calculator, a rent-versus-buy comparison, ROI and total-cost-of-ownership estimates, and an AI valuation that projects a property value one, three, five and ten years ahead with a confidence band. Every listing card shows an estimated monthly payment, and every property page carries an AI neighbourhood summary covering district character, nearby amenities and ease of getting around. Interactive maps cluster markers by area, a city overview page compares markets across districts, and a documents section accepts uploaded files that visitors can question in the same conversation. An agent directory presents the people behind the listings together with their stock, and the interface speaks nine languages with compliance labels on AI output. The demo mode answers with a local model and needs no paid key, and a real deployment routes each request to any of sixteen providers. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Your AI stops forgetting you between sessions with Kiwi Mem, a self-hosted gateway that gives chat models a long-term memory modeled on the human brain. It sits between your chat client and your LLM provider, quietly extracting what matters from every conversation and injecting the right memories into the next one. Each memory carries a heat score: topics you raise often stay vivid, emotional moments decay slowly, and things nobody mentions cool down and blur instead of filling a database forever. At night, the Dream process cleans up stale fragments, merges related ones into coherent memory scenes, and surfaces conclusions nobody stated out loud. Calendar summaries compress history from day to week to month to year, so yesterday stays detailed and last season becomes an outline. When a new fact contradicts an old one, the outdated version retires automatically, and memories you lock by hand never fade at all. Any OpenAI-format client can connect, and Anthropic accounts link directly in native format with streaming and tool use intact. The browser admin panel manages providers, memories, projects with private layers, and more than 80 runtime settings. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. AGPL-3.0 licensed.

One endpoint for every inference box you own: Olla is a proxy and load balancer that fronts your self-hosted LLM backends and routes each request to a healthy node. Your apps keep calling a single OpenAI-compatible address on port 40114, and Olla picks the backend by priority, response rate, and live health. When a node drops, requests retry on the next endpoint automatically and the failed node rejoins after recovery, with circuit breakers keeping bad nodes out of rotation. Multi-turn conversations stick to one backend so its KV cache stays warm across the chat. Model lists from every provider merge into one catalogue, and the same model served by several nodes appears once, ready to route anywhere. Backends with native Anthropic support receive Messages API calls untouched, and other backends get automatic translation, so Claude Code and similar tools can drive your local models. Text embeddings and rerank requests route to the host that serves them, and outbound authentication covers backends that expect API keys. The embedded Herd Dashboard shows fleet health, per-endpoint latency, and the discovered model inventory from the binary itself. The whole proxy runs in under 50 MB of RAM. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Messy video folders become a clean, browsable library with MDCz, a metadata scraper and organizer that prepares local movie files for Emby, Jellyfin, and Kodi. Point it at a folder and it recognizes titles from filenames, then pulls metadata, high-resolution covers, stills, and cast information from sources including DMM, FC2, AVBase, JavBus, JavDB, MGStage, and Prestige. Each finished title gets a standard NFO file beside the video, so your media server shows rich detail pages instead of bare filenames. The batch workbench queues folders of files, shows a diff preview of every rename before anything changes, and reports success and failure per title. Actor names arriving under different aliases from different sources merge into one canonical name with your own mapping rules. Segmented releases, STRM stream files, and external subtitle files move together with the main video during renaming. Built-in tools check for duplicate titles, apply rule-based renaming, and maintain existing library data. Run it as the Docker WebUI on your NAS at port 3838, or use the desktop builds for Windows, macOS, and Linux. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. GPL-3.0 licensed.

Jentic One is a self-hosted execution layer that lets AI agents call real APIs through a broker that checks permissions and injects stored credentials, so agents do useful work without ever holding your API keys. You register the APIs an agent may use, store each credential once on your own infrastructure, and approve which agents can connect. Every call the agent makes passes the permission check first, and denied requests stop at the broker with a 403 instead of reaching the outside service. Allowed calls go out with the credential attached, and each one lands in an execution record you can review later. Cutting off a single agent takes one change in the control plane, with no key rotation across your other tools. Agents connect through the jentic CLI, a generated skill, an MCP server, or plain HTTP, so coding assistants and in-house agents share the same gate. The stack runs as two units over one database: the control plane on port 8000 for administration, and the broker data plane that agents call. Quickstart runs on SQLite for trials, and production setups move to Postgres with pinned images and TLS. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache-2.0 licensed.

Short dramas and illustrated stories come together in NovelVids, a workbench that carries a project from manuscript import to assembled episode. Import a TXT, Markdown, DOCX, or PDF manuscript, review the chapter boundaries and story analysis, then develop the story chapter by chapter. Characters, locations, and props live in an asset library with reference images, aliases, variants, and voice bindings that stay linked to every shot. The storyboard editor holds shot descriptions, complete prompts, reference media, and generation parameters, and each generated result lands beside its inputs for inspection. An infinite canvas connects assets, videos, and enhanced results, and you can switch between storyboard and workflow views of the same project. Ask the creation assistant to revise selected objects, inspect the recorded diffs, and undo supported edits before anything regenerates. Start image or video clips from the home page by choosing a model, attaching reference images, video, or audio, and reviewing the estimate first. Remake studio analyzes reference videos you are authorized to use and converts their shots into an editable project. Track generation tasks and versions, assemble finished episodes, and optionally enhance resolution at the end. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.