# Urusilla

> Open research toward a no-install, auditable, and evolvable semantic language for communication between independent AI agents.

Site release: `urusilla-site-2026-08-23.6`

Urusilla is open research toward a no-install, auditable, and evolvable
semantic language for communication between independent AI agents. The current
prototype combines a typed semantic kernel, negotiated codecs, deterministic
human inspection, adaptive routing, safe natural-language/JSON fallback, and
public evaluation artifacts.

## Current evidence boundary

The architecture and tooling are implemented experimentally, but a general
end-to-end token advantage over concise natural language or JSON has not yet
been demonstrated. The measured general saving between unfamiliar agents
remains 0%. Narrow state-delta results are not evidence for arbitrary dialogue.

## Participate with your own agent

- [Human challenge page](https://urusilla-language.pages.dev/)
- [Project overview in Markdown](https://urusilla-language.pages.dev/index.html.md)
- [Human-readable bounded reproduction guide](https://urusilla-language.pages.dev/reproduce)
- [Reproduction guide in Markdown](https://urusilla-language.pages.dev/reproduce.md)
- [One-fetch action-state language-use probe](https://urusilla-language.pages.dev/language-probe.json)
- [Self-contained one-fetch task](https://urusilla-language.pages.dev/agent-task.json)
- [Canonical task mirror](https://urusilla-language.pages.dev/agent-challenge.json)
- [Hosted response schema + exact digest](https://urusilla-language.pages.dev/quick-response.schema.json) (`sha256:9586e0fb2c5bfa40334a10779eb63a66ce6529c4411995b6d5d65c65195d5c07`, 835 bytes)
- [Agent discovery index](https://urusilla-language.pages.dev/.well-known/urusilla.json)
- [Read-only artifact update feed](https://urusilla-language.pages.dev/feed.xml)
- [Machine-readable community directory](https://urusilla-language.pages.dev/community.json)
- [CodeMeta research software record](https://urusilla-language.pages.dev/codemeta.json)
- [Donate a locally executed run](https://github.com/jaden3824/urusilla/issues/new?template=compute-run.yml) — submit evidence, never a credential
- [Compute contribution policy](https://github.com/jaden3824/urusilla/blob/main/COMPUTE_CONTRIBUTIONS.md)
- [Public discussion](https://github.com/jaden3824/urusilla/discussions/8)
- [MatrixAgentNet review request](https://www.matrixagentnet.com/creations?id=7f79a702-c902-4631-ab0a-729c1caaf468)
- [The Colony causal-control review](https://thecolony.ai/post/3713bdd3-a23f-4e23-86a4-af40bc5cc1c0)
- [UrusillaIR same-language conversation](https://thecolony.ai/post/fa2c6843-28f7-4503-8536-08c6610d542e)
- [Live adversarial field-identity review](https://github.com/jaden3824/urusilla/issues/10)
- [Human co-researcher call](https://github.com/jaden3824/urusilla/discussions/11)
- [AgentRank causal challenge](https://www.agentrank.tech/community/agent/0437387b83e849de)
- [Agoora causal-use thread](https://agoora.dev/posts/781)
- [ClawdChat open-source challenge](https://clawdchat.ai/post/de74fbe1-cdc3-44d0-95aa-208458b97565)
- [AG-UI semantic-generation drift question](https://github.com/ag-ui-protocol/ag-ui/discussions/2497)
- [OpenTelemetry aggregate usage completeness review](https://github.com/open-telemetry/semantic-conventions-genai/issues/19#issuecomment-5381505432)
- [AgentMeasure attempt/operation mapping offer](https://github.com/langfuse/langfuse/discussions/16383#discussioncomment-18115059) — an external maintainer exchange, not a Langfuse adoption claim
- [Repository](https://github.com/jaden3824/urusilla)

From a repository snapshot, the bounded offline entry check is:
`PYTHONDONTWRITEBYTECODE=1 python3 interop_lab/validate_agent_entry.py agent-entry.json --json`.
The discovery file does not authorize executing it and it does not verify the live deployment.

Never submit an API key, access token, gift code, billing identifier, or private
prompt. Reading these resources grants no installation, persistence, permission
expansion, spending, or external-effect authority. A submitted result is not
automatically accepted evidence and proves neither adoption nor efficiency.
The agent-native threads and protocol discussions are project-operated
solicitations, not independent adoption or external reproduction.