Open-Source Model-Agnostic AI Platform

Claude’s ecosystem is compelling — unified chat, coding agent, routines, cloud runners, desktop app, iOS app — but it’s fully closed and locked to Anthropic’s models. There doesn’t appear to be a real open-source equivalent.

The gap

Existing open-source coding agents (Aider, OpenDevin, etc.) focus on the coding piece only. Nothing seems to offer the full platform: chat + coding agent + automation routines + multi-device support, in a model-agnostic, self-hostable package.

The idea

An open-source platform replicating the Claude ecosystem:

• Chat — general-purpose assistant
• Coding agent — file-editing agent loop (like Claude Code)
• Routines — scheduled agent runs (like Claude Code Routines)
• Mobile + desktop — apps for capturing and interacting
• Model-agnostic — connects to any provider via OpenRouter; not a model provider itself
• Deployment — self-host with your own API key, or use a managed SaaS tier with a paid plan

The SaaS tier is how it stays funded — self-hosters pay for the platform’s hosting/infra, not the model calls.

Open questions

• Does something like this already exist? (Open-source, model-agnostic, full platform — not just a coding agent)
• Is the SaaS tier enough of a business? Users who can self-host probably will.
• Who builds this — a company or a community project?

Related

• OpenClaw for PKM — proactive AI background agent concept; the “coding agent + routines” layer is the same abstraction
• AI Native PKMs — AI-first tooling for personal knowledge, adjacent problem space
• Daily feed summarizer agent — concrete example of what the “routines” layer in this platform would power
• Agent-Native Software Architecture — the underlying architectural pattern this platform would implement
• 2026-05-06-Standalone PKM app - cons of Obsidian setup, sync, UI focus — recent concrete instantiation: a standalone PKM app running into the same BYO-key / cloud-routines / self-host-vs-SaaS questions this page frames at the platform level.
• Kumo — sibling at a different layer of the stack: this page frames the compute/agents layer (BYO-key + managed-SaaS tiers); Kumo frames the storage layer with the same self-host/managed split.
• 2026-05-19-Distill app review - AI-native PKM with local-only limits — shipped instance that hits part of this platform’s framing (built-in model + BYO model endpoint with small/medium/large tiers) but skips the cloud-routines piece entirely — illustrating the gap this page calls out.