Meterless Saves AI Workflows as Reusable Missions
meterless.ai
Meterless saves your AI work as reusable 'missions' instead of tossing it after each session. Swap models freely later, reuse cuts token costs up to 90%.
Based on reporting by meterless.ai — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Most AI tools have a strange kind of amnesia. You get an answer, and then the reasoning, the plan, the memory that produced it — all of it vanishes the moment the session ends. Meterless is built around the opposite idea: that the structure behind the work, not just the output, is the thing worth keeping.
Relay is the part that automates tasks across your other applications. Describe a goal, pick which apps it can touch, and you get to inspect the plan before it runs. If it works, that run becomes a mission — something you can verify, schedule, and replay later, complete with vision-verified checkpoints and approval gates. Relay can drive up to eight real application windows per mission, which is a very different pitch from a chatbot that forgets everything the second you close the tab.
Gaia handles the memory side. It runs a twelve-brain planner over persistent, user-owned memory that's stored as plain, readable files rather than some opaque database — working, episodic, and semantic tiers you can grep, audit, and take with you. The pitch is that your projects and decisions keep compounding across sessions instead of starting from zero every single time, and that this memory stays independent of whichever model happened to generate it.
Then there's Swarms, a free browser tool where typing a single goal auto-generates an entire team of agents and runs them in parallel — up to 64 at once, part of a broader swarm capability the company says scales past 300 agents. It's open source, runs locally, and every artifact keeps its lineage so you can rerun or refine any branch of the graph later. All three products sit on the same open-source Meterless Runtime, released under Apache 2.0, which bundles four pieces: a Markovian approach to keep long tasks from blowing up context windows, a memory layer called H-MEM, shared world models so parallel agents don't drift apart, and a routing system called Scout that picks tools without needing constant prompting.
The economic pitch is straightforward: run a task once with an expensive frontier model, save the resulting mission, then replay it with cheaper models at scale for a fraction of the tokens — Meterless claims 7.3 to 15 times fewer tokens, or up to 90% less spend, depending on the workflow. Whether that holds up outside the best-case scenario is the kind of thing you'd want to test yourself, which is easier here than with most agent platforms since the engines are genuinely on GitHub, not just described in a blog post.
My take — AI-written commentary, not fact-checked reporting
The framing here is the interesting part: models are disposable, but the workflow you built around them is the actual asset, and almost nobody else is selling it that way. Keeping missions, memory, and artifacts as local, readable files instead of locking them inside one vendor's cloud is the right instinct in a market where every provider wants to be the thing you can't leave. The token-savings numbers are a range, not a guarantee, so treat them as a best case until you've run your own workflow through it — but an open-source runtime you can actually inspect is a much better trust signal than another closed agent wrapper promising efficiency.”}
Read more about this at: meterless.ai
