Cordum Packs
Official pack bundles and the public catalog for Cordum.
Ask AI about Cordum Packs
Powered by Claude Β· Grounded in docs
I know everything about Cordum Packs. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Reviews
Documentation
Cordum Packs
The official catalog of Cordum packs: versioned, production-grade bundles and runtime code that make workflows safe to operate in real systems.
Cordum is an AI agent orchestration platform with built-in brakes. It sits between agents and tools to enforce policy-before-dispatch, approval gates for risky actions, and full auditability. Packs bring those guarantees to integrations and workflows so teams can move from proof-of-concept to production with confidence.
Protocol
All packs implement the Cordum Agent Protocol (CAP) β the open wire protocol for distributed AI agent governance. Packs are CAP-native workers that communicate via BusPacket envelopes over NATS, with built-in safety gating and policy enforcement.
The reference control plane is Cordum.
From PoC to production
Most teams get stuck at the trust gap: agents can do the work, but governance, approvals, and audit trails are missing. Packs close that gap by shipping deterministic, policy-gated workflows and the runtimes that execute them. Deterministic core, autonomy at the edges.
Why teams use packs
- Policy gate for every action: approvals and constraints before execution.
- Deterministic core: predictable outcomes with replay and auditability.
- MCP-native integrations: connect tools and agents without glue code.
- Pack system extensibility: add new behaviors without changing core.
What you get
- Pack bundles: workflows, schemas, overlays, and policy fragments.
- Runtime code: workers, bridges, and receivers that execute jobs and triggers.
- Public catalog: a stable index of installable bundles.
- Security guidance: hardening and operational playbooks.
Typical outcomes
- Enrich incidents with context from tools like Slack, Jira, ServiceNow, and more.
- Route and correlate alerts across teams and systems with clear policy gates.
- Require approvals and replay execution when production changes are involved.
How packs run
A pack has two parts:
- Bundle assets (
pack/orpack.yaml): workflows, schemas, overlays, policy fragments. - Runtime code (
cmd/+internal/): workers/bridges/receivers that execute jobs or trigger workflows.
cordumctl pack install only registers the bundle assets. To execute workflows,
deploy the pack runtime(s) so the job topics they emit (for example job.slack.*)
are actually handled. Workflows can compose multiple packs, so make sure every
job topic used by a workflow has a running worker.
Install via Cordum
Set the catalog config in Cordum:
curl -X POST http://localhost:8081/api/v1/config \
-H "X-API-Key: $CORDUM_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"scope": "system",
"scope_id": "pack_catalogs",
"data": {
"catalogs": [
{
"id": "official",
"title": "Cordum Official",
"url": "https://packs.cordum.io/catalog.json",
"enabled": true
}
]
}
}'
Then open the dashboard β Packs β Marketplace.
Repo layout
cordum-packs/
packs/ # pack projects (pack.yaml at root or in pack/)
tools/ # build + catalog tooling
integrations/ # agent adapters and SDK integrations
docs/ # repo docs and roadmap notes
public/ # build output (published)
Pack projects may include worker/bridge code alongside the bundle assets. The
bundle itself lives at pack/ (or the project root if pack.yaml is there).
Build the catalog + bundles
python3 -m venv .venv
. .venv/bin/activate
pip install -r tools/requirements.txt
python tools/build.py
Output:
public/catalog.jsonpublic/packs/<id>/<version>/pack.tgz
Pack tooling
Scaffold a new pack bundle:
python tools/pack_scaffold.py my-pack \
--title "My Pack" \
--description "Pack scaffold for My Pack"
Output:
packs/my-pack/(pack bundle underpack/plus a starter README)
Audit pack bundles for required assets:
python tools/pack_audit.py
Security & operations docs
docs/security-hardening.mddocs/pack-runtime-guide.mddocs/testing-and-validation.mddocs/wiki/(wiki mirror content)
Agent adapters
Python adapters that expose MCP tools to popular agent frameworks live under
integrations/agent-adapters/. The package exports:
- LangChain tools (
build_langchain_tools) - AutoGen tools (
build_autogen_tools) - CrewAI tools (
build_crewai_tools)
Each adapter expects an MCP stdio server such as packs/mcp-bridge.
Included packs
packs/hello-pack- minimal example packpacks/mcp-bridge- MCP stdio bridge + pack bundlepacks/mcp-client- call external MCP servers with policy gatingpacks/slack- Slack ChatOps notifications and approvalspacks/github- GitHub automation workflowspacks/incident-enricher- reference pack with workers + workflows- See
packs/for the full catalog.
Publish (GitHub Pages)
The publish.yml workflow builds public/ and publishes it as GitHub Pages.
Point packs.cordum.io at the Pages domain and it will serve:
https://packs.cordum.io/catalog.json
https://packs.cordum.io/packs/<id>/<version>/pack.tgz
License
See LICENSE.
