How far can a small model go with a good harness?
We are trying harnesses that give small models more reach: short scripts that start subagents, and fine-tunes that teach them to call tools.
Small agents, and the interfaces that make them personal.
Glint Micro is an AI research lab. We make Slate, a workspace of plain files, and Opal, the agent that lives in it.
Research
Questions we think about.
How should people and agents share one workspace?
Slate and Opal are where we try ideas out: notes and apps as plain files an agent can read, and an agent that asks before it acts.
Can we simulate physical systems closely enough to learn from them?
We are starting small: hydrogen orbitals, sampled exactly and drawn live on the Opal page.
Slate
One folder. Your notes, your apps, your agent. Notes are Markdown, apps are single HTML files, automations are short scripts.
Markdown, HTML and scripts · Windows 10 & 11, Linux (.deb or AppImage), x64 · Git repos you own · No account needed
Opal
The agent built into Slate. Each capsule, a folder, gets its own Opal and one conversation that never needs a restart. It works in your files, and a command it has not been allowed waits for your OK.
Built into Slate · One conversation per capsule · Its job is a plain file · Coming soon
Lab
An independent AI research lab.
Glint Micro was founded in 2024. We research small agent technologies and the personal interfaces people use them through, and ship what we learn as software.
- Founded
- 2024
- Focus
- Small agent technologies, personal interfaces
- Also
- Simulation
- Makes
- Slate, and Opal (coming soon)
Contact
Write to us.
A question about Slate or Opal, our work, or a capsule you want to build. We read everything.