The first release, v0.1.0, is being prepared

What the private preview runs today. It is what v0.1.0 starts from, not a promise of what it will contain.

Discovery

A scan of your own private ranges, with consent recorded before the first packet, that reports its progress as it goes.

An inventory keyed on what a platform vouches for, a MAC or its own id, read from OPNsense, Portainer, Proxmox VE and Coolify.

Connecting

For each service found: which page mints its token, the least privilege that will work, and a check that calls the service before saying it worked.

Native drivers for Proxmox VE, OPNsense, Portainer and Coolify, capability manifests for a service with an HTTP API and no driver, and any MCP server over HTTP, SSE or stdio.

Models

Anthropic, OpenAI, Google, and any OpenAI-compatible endpoint, including Ollama, vLLM and LM Studio. A subagent can run on a different vendor from its parent.

Running agents

LLM agents that decide and delegate, and workflow agents that run a fixed plan with steps, parallel branches and bounded loops.

A live tree of every session and subagent, with what each is doing and its budget; spend caps; schedules on five-field cron.

Holding the risky calls

Reads run, writes wait: a destructive call is held for a person with its plan and a hash, and asked about again if the target changed.

Policies on classes of action with an auto-deny timer, two-approver policies, per-tool grants checked on every call, named accounts and an audit log.

Running it

One container and one port, with Postgres beside it, started as one Compose stack. Apache-2.0 licensed.

What it cannot do yet is written down too, in what Glion can and cannot do.