Zelkor

Use the zelkor CLI ​

The zelkor CLI packages LangGraph, Aegra, and Deep Agents projects into a zelkor-agent Helm release. It discovers the live platform (database URL, AI Gateway /v1, MCP_URL, agents Host) from Kubernetes — it does not install the platform chart.

Your agent keeps running on Zelkor’s sandbox: calls go through the AI Gateway and MCP gateway; the CLI copies platform auth and trace settings onto the worker Deployment.

Prerequisites ​

  • Python 3.10+ and pip.
  • kubectl and helm on your PATH.
  • A running Zelkor platform in a namespace you can reach.
  • For zelkor run: langgraph-sdk installed in the same environment as the CLI.
bash
git clone https://github.com/devopssquaddev/zelkor-platform.git
cd zelkor-platform
pip install -e ./cli
zelkor version

Point at a cluster ​

Named envs store only kube context and namespace — no hosts, tokens, or DSNs.

bash
zelkor env add staging \
  --kube-context my-eks \
  --namespace zelkor

zelkor env use staging
zelkor env list

Override for one command: zelkor status --env staging.
Store file: ~/.zelkor/envs.yaml or project .zelkor/env (override with ZELKOR_ENV_FILE).

zelkor dev prefers an env named local when present (kind overlay); otherwise it uses the current env.

Scaffold a project ​

bash
mkdir my-agent && cd my-agent
zelkor init

Creates agent.json and AGENTS.md. Add your graph source and optional tools.json for MCP server names.

Deploy and run ​

From the project directory (parent of charts/zelkor-agent on your machine or set ZELKOR_AGENT_CHART / --chart):

bash
zelkor deploy --env staging --registry ghcr.io/myorg
  • Non-kind clusters: --registry or ZELKOR_IMAGE_REGISTRY is required.
  • Kind: registry defaults to ghcr.io/devopssquaddev; image loads with kind load when not pushing.
  • Skip local build: ZELKOR_SKIP_BUILD=1 zelkor deploy when the image is already in the registry.
  • Platform MCP extras: if tools.json names servers not registered in workspace.tools.extraBackends, deploy fails with a GitOps snippet — fix the platform overlay first (Register extra MCP backends).

Optional workload intent (compiled to Helm values):

bash
zelkor deploy --timeout 120s --max-tokens 4096

--approval-threshold exits non-zero on Community Edition (Human-in-the-loop is Pro).

Run one Agent Protocol stream against the discovered agents host:

bash
export ZELKOR_AUTH_TOKEN="Bearer <tenant-jwt>"
zelkor run --input "Summarize tenant usage"

Optional: --url, --auth, --graph-id. Non-default graphs need X-Graph-ID; the CLI sets it when your release is not the catch-all route.

Inspect and undeploy ​

bash
zelkor status --env staging
zelkor doctor --env staging
zelkor logs --tail 100 --no-follow
zelkor undeploy

doctor checks discovered DATABASE_URL, OPENAI_BASE_URL, MCP_URL, and agents Host on the platform Aegra Deployment.

Global flags ​

FlagPurpose
--envNamed env (kube context + namespace)
--chartPath to charts/zelkor-agent
--platform-chartPath to charts/zelkor-platform (undeploy default-route restore)

Environment variables: ZELKOR_AGENT_CHART, ZELKOR_PLATFORM_CHART, ZELKOR_IMAGE_TAG, ZELKOR_DEEP_IMAGE, ZELKOR_AGENTS_URL, ZELKOR_AUTH_TOKEN.

Command reference ​

Commands registered by the zelkor binary:

CommandCEDescription
init [dir]●Write agent.json + AGENTS.md
dev [dir]●Build and helm upgrade without registry push
deploy [dir]●Build, push (or kind load), helm upgrade agent release
undeploy [dir]●helm uninstall agent; restore platform default route if needed
run [dir]●Create thread + stream run (langgraph-sdk)
logs [dir]●kubectl logs on agent Deployment (-f default)
status●List platform and agent Helm releases and HTTPRoutes
doctor●Probe discovered platform endpoints
version●CLI version; platform chart when env resolves
env add|list|use|remove●Manage kube targets
token mint●Mint RS256 JWT from cluster signing key (ops/testing)
token jwks●Write public JWKS from cluster ConfigMap
loginPro/Enterprise — exits with upgrade message on CE
licensePro/Enterprise
whoamiPro/Enterprise
teamPro/Enterprise — AITeam CRUD
budgetPro/Enterprise — AITeam.spec.budget
auditEnterprise — audit export

Paid commands print: This command requires Zelkor Pro or Enterprise. Community Edition does not apply it. and exit code 2.

token mint flags ​

--release, --tenant (required), --namespace, --kubeconfig, --context, --issuer, --audience, --ttl, --out.

token jwks flags ​

--release, --out (required), --env-file, --namespace, --kubeconfig, --context.

See also ​