Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
apiVersion: v1
kind: ServiceAccount
metadata:
name: codex-llm-proxy
namespace: control-plane
labels:
app: codex-llm-proxy
automountServiceAccountToken: false
---
apiVersion: v1
kind: ConfigMap
metadata:
name: codex-llm-proxy-config
namespace: control-plane
labels:
app: codex-llm-proxy
data:
CODEX_LLM_PROXY_RUNTIME_PORT: "8080"
CODEX_LLM_PROXY_ADMIN_PORT: "8081"
CODEX_LLM_PROXY_PROBE_PORT: "9090"
# CODEX_LLM_PROXY_MAX_BODY_BYTES is deliberately unset: the proxy derives it
# from LIMITS.maxRequestBodyBytes plus the envelope allowance (#731), and it
# governs ordinary and anonymous requests.
# V2 visual envelope. Keep this equal to LIMITS.maxVisualRequestBodyBytes.
CODEX_LLM_PROXY_MAX_VISUAL_BODY_BYTES: "25165824"
CODEX_LLM_PROXY_MAX_STREAM_DURATION_MS: "1800000"
CODEX_LLM_PROXY_MAX_DEADLINE_MS: "1800000"
CODEX_LLM_PROXY_JWT_ISSUER: control-api
CODEX_LLM_PROXY_EXECUTION_ENABLED: "false"
CODEX_LLM_PROXY_CONTROL_API_URL: http://control-api-rpc-gateway.control-plane.svc.cluster.local:8090/api/v1
CODEX_LLM_PROXY_CONTROL_API_SERVICE: codex-llm-proxy
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: codex-llm-proxy
namespace: control-plane
labels:
app: codex-llm-proxy
spec:
replicas: 1
revisionHistoryLimit: 3
selector:
matchLabels:
app: codex-llm-proxy
strategy:
type: RollingUpdate
rollingUpdate:
maxSurge: 1
maxUnavailable: 0
template:
metadata:
labels:
app: codex-llm-proxy
spec:
serviceAccountName: codex-llm-proxy
automountServiceAccountToken: false
# On SIGTERM the proxy stops accepting connections and finishes every
# request it already admitted (#739 D6, R18): queue wait 60 s + body read
# 10 s + redeem 15 s + stream cap (CODEX_LLM_PROXY_MAX_STREAM_DURATION_MS /
# 1000) + finalize 30 s (one call plus its one retry, 15 s each) + 20 s
# margin. test/deployManifest.test.ts derives this value from the proxy's
# own constants.
terminationGracePeriodSeconds: 1935
securityContext:
runAsNonRoot: true
seccompProfile:
type: RuntimeDefault
containers:
- name: codex-llm-proxy
image: clerum/codex-llm-proxy:0.1.0
imagePullPolicy: IfNotPresent
envFrom:
- configMapRef:
name: codex-llm-proxy-config
env:
- name: CODEX_LLM_PROXY_JWT_PUBLIC_KEY
valueFrom:
configMapKeyRef:
name: control-api-public-key
key: CONTROL_API_PUBLIC_KEY_PEM
- name: CODEX_LLM_PROXY_CONTROL_API_TOKEN
valueFrom:
secretKeyRef:
name: codex-llm-proxy-secrets
key: CODEX_LLM_PROXY_CONTROL_API_TOKEN
# Caps V8's old space under the 2048Mi limit so the heap is
# collected before the kernel kills the pod. Measured peak with
# eight 8 MiB streams, two 24 MiB visual streams and three queued
# 8 MiB bodies (#739 D5): 790 MiB of RSS at a 384 MiB cap, 886-1009
# MiB without one. The byte-based BodyBudget also admits about
# eight compact bodies of the worst structure the contract admits
# (#806 Q1), each a ~44 MiB tree once parsed. With 8 held streams,
# 8 queued bodies and two visual streams, 640 MiB aborted 3 of 3
# and 768 MiB passed 3 of 3; that load peaked at 1327-1460 MiB of
# RSS on macOS, which is not cgroup memory. 768 MiB is the smallest
# cap tested that passes, not a proven minimum.
- name: NODE_OPTIONS
value: "--max-old-space-size=768"
ports:
- name: runtime
containerPort: 8080
protocol: TCP
- name: admin
containerPort: 8081
protocol: TCP
- name: metrics
containerPort: 9090
protocol: TCP
readinessProbe:
httpGet: { path: /readyz, port: metrics }
initialDelaySeconds: 4
periodSeconds: 8
timeoutSeconds: 2
livenessProbe:
httpGet: { path: /healthz, port: metrics }
initialDelaySeconds: 8
periodSeconds: 12
timeoutSeconds: 2
resources:
requests:
cpu: 50m
memory: 768Mi
limits:
cpu: 250m
memory: 2048Mi
securityContext:
allowPrivilegeEscalation: false
readOnlyRootFilesystem: true
runAsNonRoot: true
runAsUser: 1001
runAsGroup: 1001
seccompProfile:
type: RuntimeDefault
capabilities:
drop: [ALL]
volumeMounts:
- name: tmp
mountPath: /tmp
volumes:
- name: tmp
emptyDir: {}
---
apiVersion: v1
kind: Service
metadata:
name: codex-llm-proxy
namespace: control-plane
labels:
app: codex-llm-proxy
spec:
type: ClusterIP
selector:
app: codex-llm-proxy
ports:
- name: runtime
port: 8080
protocol: TCP
targetPort: runtime
- name: admin
port: 8081
protocol: TCP
targetPort: admin
- name: metrics
port: 9090
protocol: TCP
targetPort: metrics