When many AI agents work a long session, each invents its own names for the same thing —
accessToken, authToken, bearer_token. By merge time the
codebase speaks five dialects of itself. SkeinRank gives one deterministic, auditable
terminology layer: agent-lexicon catches drift in your terminal,
skeinrank governs it at runtime.
pipx install agent-lexiconpipx install agent-lexicon
# first run inside your project# `alex` is the short CLI alias for `agent-lexicon`alex init alex scan README.md docs src alex review alex publish Workspace ready: lexicon/ Candidates discovered Review inbox created Snapshot publishedalex init alex scan README.md docs src alex review alex publish # Workspace ready: lexicon/ # Candidates discovered # Review inbox created # Snapshot published
The same terminology governance shows up at two moments: while agents write code, and while search, RAG, and agents run in production. You don't choose between the two projects — you grow from one into the other.
A deterministic terminology layer for AI agents — in your terminal.
pipx, initialize a workspace, and get the CLI in every projectA control plane that governs the same vocabulary in production search.
SkeinRank is not trying to replace semantic search. It gives teams a controlled language layer around it: canonical terms, aliases, evidence, owners, runtime bindings, and pinned snapshots. That is the part normal search stacks usually leave scattered across synonym files, prompts, CSVs, and tribal knowledge.
Turn internal names, acronyms, and aliases into reviewed product state instead of hidden query rewrites or one-off prompt instructions.
Bind approved language to the right search context, index, fields, filters, and snapshot so runtime systems know which interpretation is safe to use.
Surface new terminology and conflicts as evidence-backed proposals, then publish an immutable snapshot only after the change is accepted.
Use embeddings to retrieve meaning. Use SkeinRank to govern the vocabulary that shapes retrieval.
SkeinRank sits beside your existing search stack. Profiles define language, bindings select the search context, and snapshots pin the version that search, RAG, and agents can safely use.
Queries can be canonicalized and routed with an explicit binding, pinned snapshot, and policy-aware context.
Domain language becomes reviewable platform state instead of hidden synonym files or prompt-only heuristics.
Agents, search services, and CI workflows can consume the same terminology package and runtime model.
The governance console is designed as an operator workflow: run the Docker beta stack, define terminology, apply it to an index, run enrichment, audit the active snapshot, and test the runtime query path before integrating it downstream.
Create a terminology profile with canonical terms, aliases, slots, suggestions, and guardrails.
Attach one profile to one Elasticsearch index, alias, or scoped document collection.
Run dry-runs and jobs that write canonical attributes or prepare rollout metadata.
Audit the immutable runtime version currently pinned to each binding.
Verify canonical queries, matched aliases, snapshots, and search hits from the UI.
When you need more than one-off canonicalization, the governed lifecycle is already there: evidence, proposals, bindings, snapshots, rollout, and rollback.
Use bindings to decide which terminology profile, index, fields, filters, and snapshot belong to a query.
Console betaNormalize aliases and jargon into stable technical values controlled by profiles and guardrails.
AvailableValidate dictionaries, extract attributes, and canonicalize text or documents without deploying services.
AvailableReview suggestions, enforce stop-lists, manage users/tokens, and keep terminology changes auditable.
Console betaRun dry-runs and enrichment jobs against Elasticsearch bindings with job history and rollback metadata.
PreviewSee which immutable version is active for each binding and why a search context is ready or stale.
Console betaThe site presents SkeinRank honestly: the core SDK/CLI path is the smallest stable entry point, while the governance console and Elasticsearch workflows are documented as beta/platform preview.
Start with pipx install agent-lexicon for the first aha moment in your terminal, then run the full skeinrank control-plane stack when you want UI, API, bindings, evidence, and enrichment workflows.