One application, several runtimes
Flutter for web and mobile, a Dart API, and FFmpeg through WebAssembly or native code. Shared editor components handle the timeline; each device uses its own rendering engine.
Intelligent Iterations · In active development
ii-os.
One workspace for AI generation,
media editing and coding agents.
We’re building ii-os for teams that want to use AI on infrastructure they control. It connects their models, storage and compute, with shared permissions and a record of each job.
- My role
- Co-founder & Technical Lead
- Built with
- Flutter, Dart, Google Cloud
- Public release
- Architecture preview ↗
How the work moves.
Three workflows. The customer keeps the data and compute.
Generate media
- 01ii-osChoose a model and submit a job
- 02Your computeRun the generation workload
- 03Your storageSave the generated media
The app reads the result directly from the customer’s storage. ii-os coordinates the job and records its metadata.
Edit video
- 01Your mediaImport local or generated files
- 02Your deviceEdit and render with FFmpeg
- 03Your exportSave locally or to your storage
Video, audio and captions share a multi-track editor. Rendering runs in the browser or on the mobile device.
Run an agent
- 01A defined taskA person approves one issue
- 02An isolated runtimeFleet allocates compute for the agent
- 03Work to reviewFollow the result in the project
Permissions and test instructions are checked before dispatch. If the agent needs a human decision, the task returns for input.
The engineering.
Access scoped to the workspace
Each workspace gets an explicit subset of the organization’s resources. Storage access uses short-lived grants checked against the requested file.
Compute with a lifecycle
ii-fleet-runner places jobs by capability and capacity. Containers and VMs isolate the work; cleanup is verified before a machine’s capacity becomes available again.
Evaluation alongside development
LLM Lab is our related work on repeatable model evaluation: datasets, scorers, experiments and cost tracking. It gives us a way to compare changes to AI behavior.
Where it stands.
ii-os is running privately and remains in active development. We’ve published its architecture as ii-open-os; that repository is a preview, with source release still ahead.
Read the architecture on GitHub ↗