Codon SDK Monorepo

Codon SDK provides common building blocks for Codon agents, including immutable node specifications, logic ID generation, and a shared telemetry vocabulary. This repository also houses framework-specific instrumentation packages that emit OpenTelemetry spans enriched with Codon metadata.

Key Features

• Immutable NodeSpec records that introspect Python callables and generate stable SHA-256 identifiers tied to organization, role, and model metadata.
• Canonical logic request hashing that deduplicates workloads by agent class, participating nodes, and topology.
• OpenTelemetry span attribute catalog for agent runs, tools, LLMs, and vector database interactions.
• Pluggable instrumentation packages (e.g., LangGraph) that decorate nodes, capture latency, and forward spans to OTLP endpoints.

Repository Layout

README.md
sdk/
pyproject.toml # Core codon_sdk packaging metadata
src/codon_sdk/ # SDK source (schemas, instrumentation helpers)
test/ # Test scaffold
instrumentation-packages/
codon-instrumentation-langgraph/
codon/instrumentation/langgraph/ # LangGraph decorators & attributes
pyproject.toml # Package metadata
codon-instrumentation-openai/ # Placeholder for OpenAI instrumentation

Prerequisites

• Python 3.9 or newer
• pip and a virtual environment tool such as venv or pipenv
• Access to an OTLP-compatible collector if you plan to export telemetry

Installation

Clone the repository and install the core SDK in editable mode:

python -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -e sdk

Instrumentation packages are published independently. To work on one locally, install it the same way:

pip install -e instrumentation-packages/codon-instrumentation-langgraph

Note: The OpenAI package is currently a stub and will be populated in a future iteration.

Environment Configuration

Variable	Purpose
ORG_NAMESPACE	Required by NodeSpec and instrumentation to scope identifiers.
OTEL_SERVICE_NAME	Optional service name applied during telemetry initialization.
OTEL_EXPORTER_OTLP_ENDPOINT	Configure the OTLP collector when using the default exporter.

Set ORG_NAMESPACE before constructing NodeSpec objects or instrumented decorators will raise a validation error.

Working with NodeSpec

NodeSpec inspects Python callables to capture the function signature, type hints, and optional model metadata. It emits a deterministic SHA-256 ID that downstream systems can rely on.

fromcodon_sdk.instrumentation.schemas.nodespecimportNodeSpec@track_node("summarize", role="processor")defsummarize(text: str) ->str:
...
# Alternatively, construct the NodeSpec directlynodespec=NodeSpec(
org_namespace="acme",
name="summarize",
role="processor",
callable=summarize,
model_name="gpt-4o",
model_version="2024-05-13",
)
print(nodespec.id)

NodeSpec requires type annotations to build JSON schemas for inputs and outputs. If annotations are missing, the generated schemas may be empty.

Generating Logic IDs

Logic IDs canonicalize workload definitions so repeated submissions map to the same identifier.

fromcodon_sdk.instrumentation.schemas.logic_idimport (
AgentClass,
LogicRequest,
generate_logic_id,
)
logic_request=LogicRequest(
agent_class=AgentClass(
name="ReportAgent",
version="0.1.0",
description="Generates weekly status reports",
),
nodes=[nodespec],
)
logic_id=generate_logic_id(logic_request)
print(logic_id) # Stable SHA-256 hash

The hash is deterministic because nodes and topology edges are sorted prior to serialization. This enables safe retries and caching.

Instrumentation & Telemetry

Start by configuring OTEL and decorating nodes:

fromcodon.instrumentation.langgraphimportinitialize_telemetry, track_nodeinitialize_telemetry(service_name="codon-langgraph-demo")
@track_node("retrieve_docs", role="retriever")defretrieve_docs(query: str) ->list[str]:
...

When the decorated function executes, the LangGraph package:

• materializes a NodeSpec and captures its ID, signature, and schemas
• wraps execution in an OpenTelemetry span (async and sync supported)
• records inputs, outputs, and wall-clock latency via standardized span attributes

Spans are exported with org.namespace, agent.framework.name, and the Codon span names defined in codon_sdk.instrumentation.schemas.telemetry.spans.

Development Workflow

• Run formatting/linting aligned with your team preferences (no toolchain is enforced yet).
• Execute tests with pytest once suites are populated:

  pytest sdk/test

• Use pip install -e <package> for editable installs while iterating.

Roadmap & Known Gaps

• codon-instrumentation-openai awaits implementation.
• Minimal test coverage is checked in; broaden it as modules stabilize.
• README content inside codon_sdk will be expanded with module-level documentation.

License

Package metadata references the MIT License. Add the license file before publishing to PyPI.