Chat & streaming
Native chat with multi-modal input, tool-call rendering, approvals, session archive, and 28 built-in slash commands.
Ai Keeper is the native macOS control surface for local AI. Serve LLMs on Apple Silicon, orchestrate multi-agent workspaces, and reach 20 channels — all from one app. Private by default, open when you need it.
Serve models, orchestrate work, extend with tools. Each layer stays inspectable — see what is running instead of guessing where a feature lives.
Switch between vllm-mlx, mlx-lm, vmlx, and llama.cpp. Publish an OpenAI- and Ollama-compatible API with SSE streaming, tool injection, and MCP JSON-RPC from a single proxy on port 8000.
Coordinate multi-agent workspaces with sequential or parallel execution, delegation jobs, standing orders, hooks, and scheduled task flows. Approvals and a real-time WebSocket control plane keep state in view.
Mount MCP servers, ingest documents into RAG, drop plugins from ClawHub, ship to Slack or WhatsApp, or hand work to 15 cloud providers with automatic failover. Mix what you need per task.
Every surface is visible — chat, agents, canvas, browser, MCP, RAG, scheduling. Navigate the product the way you navigate the work.
Native chat with multi-modal input, tool-call rendering, approvals, session archive, and 28 built-in slash commands.
Compose agents with distinct tools, hooks, and standing orders. Run them sequentially, in parallel, or as delegation jobs.
Push live HTML, CSS, and JS from agents. Watch files on disk, replay history, and eval results in place.
Drive real browsers with 14 dedicated tools — navigation, DOM inspection, screenshots, and form input.
Ingest PDFs and text, build embeddings, and query your own knowledge store from any chat or agent run.
Connect to external MCP servers as a client, or expose Ai Keeper's own tools as an MCP server to other agents.
Compose cron-driven task flows, standing orders, hooks, and background jobs. Approve sensitive steps before they run.
Generate images, video, and music. Run ComfyUI workflows, start voice calls, wire up audio channels.
Secrets vault, content moderation, request inspection, GPU and perf gauges, logs, retry policies, rate limits.
Route a single conversation across 15 cloud providers with failover, or serve the same agents into 20 chat platforms. Pull models from HuggingFace, packages from ClawHub, capabilities from MCP.
A grounded operator manual — every screen, every setting, every recipe. Each row pairs a plain-English line with the precise operator detail behind it. Search across all pages with ⌘K, look up a term in the floating glossary.
What Ai Keeper does, the first-run order, and a quick concept dictionary.
→ 02What every hub, tab, and advanced system panel is for.
→ 03What each setting changes, when to touch it, and what to verify after.
→ 04Run a model, expose the API, add knowledge, build agents, automate, recover.
→ 05Server vs Client mode, the four registries, identity, presence, protocols.
→Free. Native. Built for Apple Silicon. No telemetry, no account, no cloud by default.