Official first-party runtime · v0.1.16

Run governed AI agents locally with cryptographic proof.

Loom is Meridian's flagship runtime product. Install in one command, create a governed personal agent, run a persistent local loop, and inspect the receipts instead of guessing what the agent actually did.

Loom is the product front door now. Kernel remains runtime-neutral. First-party Meridian apps remain valuable. But if someone asks what to install instead of another local agent runtime, the answer is Loom.

Install

1 command

Binary-first installer with source fallback only when needed.

Create

new-agent

Provision a governed personal agent and wire it into Kernel.

Run

run-agent

Start a persistent heartbeat-driven local agent loop.

Prove

PoGE

Inspect receipts, audit paths, memory events, and parity views.

Live runtime snapshot

These numbers come from the live Loom-backed host. They do not replace local install proof, but they show the runtime surface is not hypothetical.

Runtime

Footprint now

Loading runtime footprint…

Queue

Pressure now

Loading queue state…

Proof

Boundary now

Loading proof status…

Install Loom now

This is the one command. Everything else on this page assumes you started here.

One-command install

curl -fsSL https://raw.githubusercontent.com/mapleleaflatte03/meridian-loom/main/scripts/install.sh | bash

The installer prefers a matching GitHub release asset, provisions $HOME/.local/share/meridian-loom, and ends by telling you to run loom new-agent.

Create your first governed personal agent

Initialize Kernel

Run python3 quickstart.py --init-only inside /opt/meridian-kernel.

Provision agent

loom new-agent registers the agent and creates the local folder.

Run loop

loom run-agent starts the persistent heartbeat-driven local process.

Inspect proof

Use loom status, loom channel deliveries, and loom memory search.

export LOOM_ROOT="${HOME}/.local/share/meridian-loom/runtime/default"
export MERIDIAN_KERNEL_PATH=/opt/meridian-kernel
export MERIDIAN_ORG_ID="${MERIDIAN_ORG_ID:-local_foundry}"

cd /opt/meridian-kernel
python3 quickstart.py --init-only

loom new-agent \
  --name "My Assistant" \
  --root "$LOOM_ROOT" \
  --kernel-path "$MERIDIAN_KERNEL_PATH" \
  --org-id "$MERIDIAN_ORG_ID"

loom run-agent my-assistant

What Loom ships right now

Personal agents

Real agent folders, runtime profiles, heartbeat loops, and Kernel registration instead of a vague assistant prompt.

Terminal and browser jobs

Bounded terminal execution and browser navigation are already live surfaces with receipts and audit artifacts.

Channels and memory

Telegram and webhook delivery can be wired into the generated agent config, and memory operations now emit receipts.

Doctor and status

First-run commands are built to tell an operator what to do next instead of forcing a source dive.

Why Loom instead of the field

Loom is not claiming to beat the field on raw breadth. It is trying to become the runtime people choose when governance and inspectable proof matter enough to change the buying decision.

Loom edge

Where the runtime is strongest

governance

9.6

proof

9.0

packaging

5.8

breadth

4.4

Named rivals

What Loom still has to beat

01OpenClaw / OpenHands still own surface breadth and broader user pull.

02OpenFang still owns the most compact runtime packaging story.

03TEMM1E / Goose / IronClaw still feel stronger on persistent local agency and memory density.

04CrewAI / LangGraph still dominate orchestration mindshare.

Read the full compare page →

Need the next step?

Open Loom on GitHub See first-party apps on Loom Read the boundary note