Blog
Engineering notes.
Notes on verifiability, measurement, and the serving stack.
-
6 October 2026 · Agent infrastructure
Captain Code adds OpenShell: bounded agents, inspectable work
Scoped sandbox execution today. A path toward proof-backed audit evidence and testable replay, with the limits stated.
-
23 September 2026 · Open source
Captain Code is open source
One terminal for every coding agent you already pay for. Captain Code picks a model for each task, hands the work on when one hits its limit, and puts open models to work when they are good enough. MIT, local, no telemetry.
-
10 September 2026 · Tender compliance
Tender integrity needs proof, not another audit log
High-stakes sealed tenders still rest on operator trust. An integrity layer under the sourcing platform can prove the process without replacing it.
-
2 September 2026 · Agentic systems in regulated environments
Deterministic AI for regulated systems
How critical teams should reason about replay, traceability, and the decision gap between model quality and deterministic behavior.
-
1 September 2026 · Agentic Determinism Index (ADI)
Yes, LLMs can be deterministic. But most hosted stacks are not.
The folk explanation for LLM variance blames the model. The folk explanation is wrong. An open harness measures what hosted stacks actually do. The live index keeps scoring hosted setups for byte-exact replay.