Observability Dashboard Reference

This reference covers the Helm chart values, HTTP and WebSocket endpoints, WebSocket event types, and configuration environment variables for the Observability Dashboard.

Helm chart values

Field Type Default Description

replicaCount

integer

1

Number of dashboard replicas.

image.repository

string

ghcr.io/agentic-layer/observability-dashboard

Container image repository.

image.tag

string

"" (chart appVersion)

Container image tag. Empty string uses the chart’s appVersion.

image.pullPolicy

string

IfNotPresent

Kubernetes image pull policy.

namespace.create

boolean

true

When true, the chart creates the target namespace if it does not exist.

namespace.name

string

observability-dashboard

Namespace into which the dashboard is deployed.

service.type

string

ClusterIP

Kubernetes Service type.

service.port

integer

8000

Port exposed by the Service.

containerPort

integer

8000

Port the container listens on.

env

list

[{name: LOGLEVEL, value: INFO}]

Environment variables injected into the container. See Configuration environment variables.

readinessProbe.initialDelaySeconds

integer

5

Seconds to wait before the first readiness check.

readinessProbe.periodSeconds

integer

5

Interval between readiness checks.

readinessProbe.httpGet.path

string

/health

Path used for the readiness HTTP probe.

readinessProbe.httpGet.port

integer

8000

Port used for the readiness HTTP probe.

livenessProbe.initialDelaySeconds

integer

30

Seconds to wait before the first liveness check.

livenessProbe.periodSeconds

integer

30

Interval between liveness checks.

livenessProbe.httpGet.path

string

/health

Path used for the liveness HTTP probe.

livenessProbe.httpGet.port

integer

8000

Port used for the liveness HTTP probe.

resources.limits.cpu

string

100m

CPU limit for the container.

resources.limits.memory

string

128Mi

Memory limit for the container.

resources.requests.cpu

string

100m

CPU request for the container.

resources.requests.memory

string

128Mi

Memory request for the container.

HTTP endpoints

Method Path Description

POST

/v1/traces

OTLP/HTTP trace ingest. Accepts application/x-protobuf or application/json content types. Handles optional gzip compression (Content-Encoding: gzip). Returns 200 OK on success with a serialized ExportTraceServiceResponse. Returns 400 for invalid gzip or malformed protobuf, 415 for unsupported content types.

GET

/health

Liveness and readiness probe. Returns {"status": "healthy"} with HTTP 200 when the service is running.

GET

/api/filters

Returns active conversation IDs and workforce names seen in the last 24 hours. Response shape: {"conversation_ids": [...], "workforce_names": [...]}.

GET

/api/filters/stats

Returns counts of currently tracked filter values. Response shape: {"conversation_ids_count": N, "workforce_names_count": N}.

GET

/

Serves the bundled single-page application (SPA) at the root path and all unmatched paths via the index.html fallback.

WebSocket endpoints

Path Description

/ws

Global event stream. Receives all processed events from all agents. Accepts optional query parameters conversation_id and workforce to filter events server-side (see WebSocket filtering).

/ws?conversation_id={uuid}

Filtered stream. Only events where conversation_id matches the given UUID4 value are sent to this connection.

/ws?workforce={name}

Filtered stream. Only events where the agentic_layer.workforce resource attribute matches the given workforce name are sent to this connection.

On connection, the server sends a welcome message:

{
  "type": "connection_established",
  "message": "Connected to observability dashboard",
  "filters": {
    "conversation_id": null,
    "workforce": null
  }
}

WebSocket filtering

Both filter parameters can be combined: /ws?conversation_id={uuid}&workforce={name}. An event is delivered only when all specified non-null filters match.

conversation_id must be a valid UUID4 string (36 characters, format ^[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$).

WebSocket event types

All events share the BaseEvent fields. Additional fields are present for specific event types.

Base event fields

Field Type Description

event_type

string

Discriminator. One of the values listed in the table below.

acting_agent

string

Name of the agent that generated this event.

conversation_id

string

UUID4 identifier for the conversation this event belongs to.

timestamp

string

ISO 8601 UTC timestamp (e.g. 2024-01-15T10:30:00Z).

invocation_id

string

Identifier for one agent invocation within a conversation. Empty string if not set.

workforce_name

string or null

Value of the agentic_layer.workforce OTel resource attribute, or null if not present.

Event type reference

event_type Description Additional fields

agent_start

Agent begins execution.

None beyond base fields.

agent_end

Agent completes execution.

None beyond base fields.

llm_call_start

LLM request initiated.

model (string): model name.
content (LlmRequestContent): role and content of the request.

llm_call_end

LLM response received.

content (LlmResponseContent): role and response parts.
usage_metadata (UsageMetadata): token usage breakdown.

llm_call_error

LLM call failed.

model (string): model name.
content (LlmRequestContent): the request that failed.
error (string): error message.

tool_call_start

Tool invocation begins.

tool_call (ToolCall): tool_name and arguments.

tool_call_end

Tool invocation completes.

tool_call (ToolCall): tool_name and arguments.
response (object): tool output.

tool_call_error

Tool invocation failed.

tool_call (ToolCall): tool_name and arguments.
error (string): error message.

invoke_agent_start

Agent-to-agent call begins.

tool_call (ToolCall): the tool call that triggered the invocation.
invoked_agent (string): name of the agent being called.

invoke_agent_end

Agent-to-agent call completes.

tool_call (ToolCall): the tool call that triggered the invocation.
response (object): the called agent’s output.
invoked_agent (string): name of the agent that was called.

Configuration environment variables

Variable Default Set in Description

LOGLEVEL

INFO

values.yaml env

Controls the Python logging level for the application. Accepted values: DEBUG, INFO, WARNING, ERROR, CRITICAL.

Related