See what changed and why.
Capture goals, observations, evidence, decisions, and actions in one compact artifact so teammates can review an agent's work without replaying the whole session.
About
Why this, why now.
As agents move from demos into real workflows, their work needs to remain reviewable after the chat ends. Scholialang turns goals, evidence, decisions, actions, and conclusions into a structured trace that humans can read and tools can validate.
Why use Scholialang
Reasoning traces that survive real work.
Move reasoning between tools.
Give agent hosts, workflow runners, CI jobs, and review systems a common trace format so context follows the work instead of staying locked in one chat or log.
Check claims before trusting results.
Typed evidence, stable references, confidence, criticality, and contradictions make it easier to compare traces, spot weak assumptions, and ask for targeted review.
Make automated decisions inspectable.
As AI systems touch code, operations, research, and business decisions, teams need artifacts they can diff, sign, validate, and discuss without exposing hidden chain-of-thought.
Authors + license
Who made this.
License posture: MIT OR Apache-2.0, at your option.
Affiliation: Doug Fir Labs — a neurosymbolic cognitive architecture and autonomous-runbook framework.
Contribute
Feedback welcome.
The language split is now tracked across three public repos:
scholialang-spec
for the specification and examples,
scholialang
for the Python reference package, and
scholialang-mcp
for MCP and LSP protocol tooling. File spec issues in
scholialang-spec; implementation issues belong in the
package or protocol repo.
Release
Initial v0.6 release.
-
v0.6 · content-addressable substrate
The initial public v0.6 release defines the 32-atom catalog, content-addressable
canonical_id, acanonical_id-keyed DAG registry overREFER/IMPLIESedges, and a lazy canonical-prelude with three core modes:hash_only,hash_list, andinline. What's new in v0.6 →