Configuration Reference

The following is a complete reference of the plano_config.yml that controls the behavior of a single instance of the Plano gateway. This where you enable capabilities like routing to upstream LLm providers, defining prompt_targets where prompts get routed to, apply guardrails, and enable critical agent observability features.

Plano Configuration - Full Reference
  1# Plano Gateway configuration version
  2version: v0.4.0
  3
  4# External HTTP agents - API type is controlled by request path (/v1/responses, /v1/messages, /v1/chat/completions)
  5agents:
  6  - id: weather_agent # Example agent for weather
  7    url: http://localhost:10510
  8
  9  - id: flight_agent # Example agent for flights
 10    url: http://localhost:10520
 11
 12# MCP filters applied to requests/responses (e.g., input validation, query rewriting)
 13filters:
 14  - id: input_guards # Example filter for input validation
 15    url: http://localhost:10500
 16    # type: mcp (default)
 17    # transport: streamable-http (default)
 18    # tool: input_guards (default - same as filter id)
 19
 20# LLM provider configurations with API keys and model routing
 21model_providers:
 22  - model: openai/gpt-4o
 23    access_key: $OPENAI_API_KEY
 24    default: true
 25
 26  - model: openai/gpt-4o-mini
 27    access_key: $OPENAI_API_KEY
 28
 29  - model: anthropic/claude-sonnet-4-0
 30    access_key: $ANTHROPIC_API_KEY
 31
 32  - model: mistral/ministral-3b-latest
 33    access_key: $MISTRAL_API_KEY
 34
 35  - model: groq/llama-3.3-70b-versatile
 36    access_key: $GROQ_API_KEY
 37
 38  # passthrough_auth: forwards the client's Authorization header upstream instead of
 39  # using the configured access_key. Useful for LiteLLM or similar proxy setups.
 40  - model: openai/gpt-4o-litellm
 41    base_url: https://litellm.example.com
 42    passthrough_auth: true
 43
 44  # Custom/self-hosted endpoint with explicit http_host override
 45  - model: openai/llama-3.3-70b
 46    base_url: https://api.custom-provider.com
 47    http_host: api.custom-provider.com
 48    access_key: $CUSTOM_API_KEY
 49
 50# Model aliases - use friendly names instead of full provider model names
 51model_aliases:
 52  fast-llm:
 53    target: gpt-4o-mini
 54
 55  smart-llm:
 56    target: gpt-4o
 57
 58# routing_preferences: top-level list that tags named task categories with an
 59# ordered pool of candidate models. Plano's LLM router matches incoming requests
 60# against these descriptions and returns an ordered list of models; the client
 61# uses models[0] as primary and retries with models[1], models[2]... on 429/5xx.
 62# Requires overrides.llm_routing_model to point at Plano-Orchestrator (or equivalent).
 63# Each model in `models` must be declared in model_providers above.
 64# selection_policy is optional: {prefer: cheapest|fastest|none} lets the router
 65# reorder candidates using live cost/latency data from model_metrics_sources.
 66routing_preferences:
 67  - name: code generation
 68    description: generating new code snippets, functions, or boilerplate based on user prompts or requirements
 69    models:
 70      - anthropic/claude-sonnet-4-0
 71      - openai/gpt-4o
 72      - groq/llama-3.3-70b-versatile
 73  - name: code review
 74    description: reviewing, analyzing, and suggesting improvements to existing code
 75    models:
 76      - anthropic/claude-sonnet-4-0
 77      - groq/llama-3.3-70b-versatile
 78    selection_policy:
 79      prefer: cheapest
 80
 81# HTTP listeners - entry points for agent routing, prompt targets, and direct LLM access
 82listeners:
 83  # Agent listener for routing requests to multiple agents
 84  - type: agent
 85    name: travel_booking_service
 86    port: 8001
 87    router: plano_orchestrator_v1
 88    address: 0.0.0.0
 89    agents:
 90      - id: rag_agent
 91        description: virtual assistant for retrieval augmented generation tasks
 92        input_filters:
 93          - input_guards
 94
 95  # Model listener for direct LLM access
 96  - type: model
 97    name: model_1
 98    address: 0.0.0.0
 99    port: 12000
100    timeout: 30s          # Request timeout (e.g. "30s", "60s")
101    max_retries: 3        # Number of retries on upstream failure
102    input_filters:        # Filters applied before forwarding to LLM
103      - input_guards
104    output_filters:       # Filters applied to LLM responses before returning to client
105      - input_guards
106
107  # Prompt listener for function calling (for prompt_targets)
108  - type: prompt
109    name: prompt_function_listener
110    address: 0.0.0.0
111    port: 10000
112
113# Reusable service endpoints
114endpoints:
115  app_server:
116    endpoint: 127.0.0.1:80
117    connect_timeout: 0.005s
118    protocol: http        # http or https
119
120  mistral_local:
121    endpoint: 127.0.0.1:8001
122
123  secure_service:
124    endpoint: api.example.com:443
125    protocol: https
126    http_host: api.example.com  # Override the Host header sent upstream
127
128# Optional top-level system prompt applied to all prompt_targets
129system_prompt: |
130  You are a helpful assistant. Always respond concisely and accurately.
131
132# Prompt targets for function calling and API orchestration
133prompt_targets:
134  - name: get_current_weather
135    description: Get current weather at a location.
136    parameters:
137      - name: location
138        description: The location to get the weather for
139        required: true
140        type: string
141        format: City, State
142      - name: days
143        description: the number of days for the request
144        required: true
145        type: int
146    endpoint:
147      name: app_server
148      path: /weather
149      http_method: POST
150    # Per-target system prompt (overrides top-level system_prompt for this target)
151    system_prompt: You are a weather expert. Provide accurate and concise weather information.
152    # auto_llm_dispatch_on_response: when true, the LLM is called again with the
153    # function response to produce a final natural-language answer for the user
154    auto_llm_dispatch_on_response: true
155
156# Rate limits - control token usage per model and request selector
157ratelimits:
158  - model: openai/gpt-4o
159    selector:
160      key: x-user-id       # HTTP header key used to identify the rate-limit subject
161      value: "*"           # Wildcard matches any value; use a specific string to target one
162    limit:
163      tokens: 100000       # Maximum tokens allowed in the given time unit
164      unit: hour           # Time unit: "minute", "hour", or "day"
165
166  - model: openai/gpt-4o-mini
167    selector:
168      key: x-org-id
169      value: acme-corp
170    limit:
171      tokens: 500000
172      unit: day
173
174# Global behavior overrides
175overrides:
176  # Threshold for routing a request to a prompt_target (0.0–1.0). Lower = more permissive.
177  prompt_target_intent_matching_threshold: 0.7
178  # Trim conversation history to fit within the model's context window
179  optimize_context_window: true
180  # Use Plano's agent orchestrator for multi-agent request routing
181  use_agent_orchestrator: false
182  # Connect timeout for upstream provider clusters (e.g., "5s", "10s"). Default: "5s"
183  upstream_connect_timeout: 10s
184  # Path to the trusted CA bundle for upstream TLS verification
185  upstream_tls_ca_path: /etc/ssl/certs/ca-certificates.crt
186  # Model used for intent-based LLM routing (must be listed in model_providers)
187  llm_routing_model: Plano-Orchestrator
188  # Model used for agent orchestration (must be listed in model_providers)
189  agent_orchestration_model: Plano-Orchestrator
190  # Disable agentic signal analysis (frustration, repetition, escalation, etc.)
191  # on LLM responses to save CPU. Default: false.
192  disable_signals: false
193
194# Model affinity — pin routing decisions for agentic loops
195routing:
196  session_ttl_seconds: 600    # How long a pinned session lasts (default: 600s / 10 min)
197  session_max_entries: 10000  # Max cached sessions before eviction (upper limit: 10000)
198  # session_cache controls the backend used to store affinity state.
199  # "memory" (default) is in-process and works for single-instance deployments.
200  # "redis" shares state across replicas — required for multi-replica / Kubernetes setups.
201  session_cache:
202    type: memory              # "memory" (default) or "redis"
203    # url is required when type is "redis". Supports redis:// and rediss:// (TLS).
204    # url: redis://localhost:6379
205    # tenant_header: x-org-id  # optional; when set, keys are scoped as plano:affinity:{tenant_id}:{session_id}
206
207# State storage for multi-turn conversation history
208state_storage:
209  type: memory            # "memory" (in-process) or "postgres" (persistent)
210  # connection_string is required when type is postgres.
211  # Supports environment variable substitution: $VAR or ${VAR}
212  # connection_string: postgresql://user:$DB_PASS@localhost:5432/plano
213
214# Input guardrails applied globally to all incoming requests
215prompt_guards:
216  input_guards:
217    jailbreak:
218      on_exception:
219        message: "I'm sorry, I can't help with that request."
220
221# OpenTelemetry tracing configuration
222tracing:
223  # Random sampling percentage (1-100)
224  random_sampling: 100
225  # Include internal Plano spans in traces
226  trace_arch_internal: false
227  # gRPC endpoint for OpenTelemetry collector (e.g., Jaeger, Tempo)
228  opentracing_grpc_endpoint: http://localhost:4317
229  span_attributes:
230    # Propagate request headers whose names start with these prefixes as span attributes
231    header_prefixes:
232      - x-user-
233      - x-org-
234    # Static key/value pairs added to every span
235    static:
236      environment: production
237      service.team: platform