An engraved moon with shaded craters above an etched landscape.

Accelerating Uncertain computation

Ocuna builds the execution layer for AI workloads that branch as they run. From model training to agent research, we’re building systems that coordinate work and keep their decisions on the record.

Supported By

Compute follows the work.

Ocura directs compute across an execution frontier. The scheduler places work, holds budgets, and reuses results across training and inference branches. You build the pipeline; the runtime governs what runs, what waits, and what continues.

Training

Explore more than one path.

Coordinate checkpoint, data, and parameter branches. Keep each run tied to its budget and the evidence behind its continuation.

Explore the execution graph

Inference

Direct a changing frontier.

Coordinate serving pools, agent endpoints, and tools under explicit policy. Keep escalation, preemption, and continuation traceable.

Explore the execution graph

Run records

Keep the decisions with the result.

Ocura OSS records local command runs and their lineage. Branch from a baseline, compare outcomes, and inspect the record behind a result.

Read about run records

Execution graph

The normal path writes hello_world. The review branch writes review branch, with both paths connected to the same chokepoint.

Record, branch, and compare.

Ocura OSS 0.5.0 is the part you can run on your own machine today. Keep a baseline, branch from it with a reason, and connect each result to the original run through the JSON CLI, the Python API, or an AI agent.

python -m pip install ocura-oss

Documentation Try the training example