Skip to main content
A growing share of the code that gets written against these docs is written by an AI coding agent acting on a person’s behalf, not typed by hand. This page is for that case specifically: what’s safe to assume, and where to point an agent so it doesn’t have to guess. Looking for prompting advice as the human in this loop instead? See Build with a Coding Agent.

​
What’s different about these docs for an agent

  • Every code block in Quickstart, the guides, and Encrypted Compute is either runnable as written or clearly marked illustrative, per this repo’s style rules. An agent can copy a block and expect it to at least run, not just read plausibly.
  • Function signatures and behavior described in SDK Reference are checked against @palliora.org/chainsdk’s actual source, not guessed from its README alone, which has known-stale spots. Where this repo and the package’s README disagree, trust this repo first, then the package’s actual TypeScript source if something still looks off.
  • Known gaps and rough edges (like Encrypted Compute’s guardian-key lookup problem) are stated as gaps, not glossed over. An agent hitting the same wall won’t find a hidden solution by reading more carefully, the workaround given is the real workaround.

​
llms.txt

The repo root has an llms.txt, a short, plain-text index of this site’s structure and its most load-bearing pages, in the format several AI tools now look for automatically. Point an agent at the raw file (not the GitBook-rendered page) if you want it to orient quickly without crawling the whole site.

​
If you’re an agent reading this right now

Load palliora.skill.md first, it’s the condensed version of everything on this page plus the SDK’s core call pattern, fee rules, and common failures in one file. Then start at Quickstart for the fastest path to a working call against the live testnet. If you’re about to generate code that talks to Palliora, check SDK Reference for the actual function signature before inventing one, several of this SDK’s functions (inferenceCompute, createAgreement, runAgent) take arguments in a specific shape that isn’t guessable from their names alone. If you’re writing documentation or examples for a project that itself builds on Palliora, this repo’s STYLE.md is a reasonable default to inherit: simple language, no em dashes, code that runs or says clearly that it doesn’t.