Roadmap

OpenLLM today is a single-user gateway over your own accounts. Here is the direction it grows in, grouped by theme. Order and timing can shift, but this is what is being built toward.

Directional, not dated

This page describes intent, not shipped features. Everything on it is subject to change, and nothing here should factor into what you buy today — the Pricing page describes the product as it exists now.

Smarter core

Auto model. One auto model that picks the right model per task: fast for simple turns, heavier when the work demands it. Where a fallback chain is an order you set, auto decides per request.
Smart routing. Per-provider caps plus latency-aware fallback that reroutes around a congested provider in real time and returns the moment it recovers.

Control and governance

Per-key model whitelist. Lock any sk-llm-… key to an explicit set of models, so you can hand a key to an agent or a teammate knowing it can only call what you allow.
Per-model token caps. Set a token or dollar budget per model, per key, per day.

These controls are also the groundwork for team and enterprise use.

Collaboration

Teams. Invite your team and share memory and private indexes across the group, without ever sharing credentials. OpenLLM is single-user today; this is the path out of that.
Git-native shared indexes. Codebase indexing built on Git: the team shares a verified baseline index, and each member only indexes their own diff.

Ecosystem

Plugins, skills, and setups. One-line installs across Claude Code, Codex CLI, Gemini CLI, Kimi, Aider, and beyond, plus a plugin marketplace.

New surfaces

Hosted agents. Run agents like Hermes and OpenClaw hosted on OpenLLM, plugged straight into your memory, indexes, and fallback chains.

Pricing

What the product is and costs today.

Core concepts

Tiers, providers, and how routing works now.