> ## Documentation Index > Fetch the complete documentation index at: https://unmute.ai/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > Unmute compiles to exactly four targets. Pipecat and LiveKit are code targets: compile writes a Python project you run. SLNG is a hosted target: compile writes a deployment body and SLNG runs the agent, so it has no `unmute dev`. Twilio compiles to a small Python app that Twilio ConversationRelay calls; you host it, and it has no `unmute dev` either. Those four are the only values `provider` accepts in `targets.yaml`. Deepgram and ElevenLabs appear in these docs as model vendors, which is not the same thing as a target, and `slng` is both. > The Go structs in `internal/spec` and `internal/ir` are the schema truth. Check a field against them, or run `unmute validate`, rather than against what you remember. # What the model reads in a task > What a step hears of the call, the small results a task sends to the model, and the limits on placeholder names. A task answers the model with a few small results, and a step opens on a chosen part of the call so far. This page lists both. ## Quickstart ```python agent.py theme={null} from unmute_core.tasks import REFUSED_BUSY, Refused, ToolDeclaration, ToolStatus Refused(refused="Run take_order first.").model_dump() # {"refused": "Run take_order first."} ToolStatus(status="completed").model_dump() # {"status": "completed"} ``` ## Why use unmute for this The results are small and frozen, and `unmute_core.tasks` names them. Your code can build the same JSON the adapters send, so a test or a custom host reads what the model reads. ## What the model reads when a task runs | Name | What it is | | - | - | | `Refused` | The result of a refused call: one sentence that says why and what to do. | | `ToolStatus` | A result that says how a call stands, like `completed` or `not_run_yet`. | | `ToolDeclaration` | A tool as the model reads it: `name`, `description` and `parameters`. `to_dict()` gives the dict a model call takes. Its parameters are a copy, so editing one declaration never changes another. | | `REFUSED_BUSY` | The result the model reads when it starts a task while another one runs. It is a `Refused` as a dict. | All of them are frozen. An unknown field raises a `ValidationError`. ## What a step hears of the call `TaskSpec.history` sets what a step opens on: `messages` (default), `full`, `last_n`, `summary` or `reset`. A handoff takes the same values. [Handoffs](/core/build/handoffs#what-the-next-agent-hears) explains each one. * `include_tool_calls: false` leaves tool calls and results out of `full` and `last_n`. * `last_n` needs `max_messages`, from 1 to 200, and any other history refuses it. * A `summary` step asks for its summary as it opens. A summary that fails gives the step `messages`. * In an `isolated` group, every step starts fresh, whatever its `history`. ```python theme={null} TaskSpec(name="confirm_order", instructions="...", history="last_n", max_messages=6) ``` ## Placeholder names A plan or a spec that does not fit raises `ValueError` when it is built. A native model call has no limit on a placeholder name. A model call that fills the prompt on its own side can name a limit in `variable_key_max`. The SLNG Context Router names 64. The check runs at build, against the call's model. 1. A placeholder name becomes a template-variable key. `a.b` is `a__b`. 2. When the model call names a limit, a key over it is refused, with a message that names the placeholder. 3. The check covers a task's instructions, and the agent's main prompt when the adapter builds the agent. 4. A model call with no `variable_key_max`, like a framework's own LLM plugin, checks nothing. ## Troubleshooting ### A placeholder name is refused when the agent is built The model call names a `variable_key_max`, and the key of a placeholder is longer. **Fix:** shorten the State value name that the placeholder reads. `a.b` counts as `a__b`. In the salon tutorial: [Book step by step](/core/tutorial/book-step-by-step). Every argument: [Tasks reference](/core/reference/tasks#taskspec). ## Where to go next What the next agent hears when a step hands off. Define a task and check it with TaskPlan. The model call that names a placeholder limit. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.