Self-improving AI Employees for everyone
Python ai-agentsai-employeeai-employeesassistantautonomous-agentsco-employeescompanygrok-botinstinctloop-agentsmuseown-your-datapersonalself-hosted
"""Perplexity Sonar Search — Wave 11.C migration.
AI-powered search returning markdown answer + citations. The
Connection facade injects ``Authorization: Bearer <api_key>`` via
``ApiKeyCredential.key_location="bearer"``. Single-op declarative-ish:
imperative body is short enough that a routing DSL roundtrip would
add more lines than it saves.
"""
from __future__ import annotations
from typing import List, Literal, Optional
from pydantic import BaseModel, Field
from services.plugin import (
ActionNode,
ApiKeyCredential,
NodeContext,
Operation,
ProbeResult,
TaskQueue,
)
class PerplexityCredential(ApiKeyCredential):
id = "perplexity"
display_name = "Perplexity"
category = "Search"
key_name = "Authorization"
key_location = "bearer"
docs_url = "https://docs.perplexity.ai/guides/getting-started"
@classmethod
async def _probe(cls, api_key: str) -> ProbeResult:
# Perplexity exposes no cheap auth-gated endpoint: /v1/models
# returns 200 for any key (including garbage), and the chat /
# search / embeddings endpoints all charge tokens per call.
# Store the key as-is; runtime calls will surface 401 if the
# key is wrong.
return ProbeResult(valid=True)
class PerplexityResult(BaseModel):
url: str = ""
class PerplexitySearchOutput(BaseModel):
query: str
answer: str = ""
citations: List[str] = Field(default_factory=list)
results: List[PerplexityResult] = Field(default_factory=list)
images: Optional[List[dict]] = None
related_questions: Optional[List[str]] = None
model: str = "sonar"
provider: Literal["perplexity"] = "perplexity"
class PerplexitySearchParams(BaseModel):
tool_name: str = Field(
default="perplexity_search",
description="Override name shown to the LLM when used as a tool.",
)
tool_description: str = Field(
default="AI-powered search using Perplexity Sonar. Returns a markdown-formatted answer with inline citation references and source URLs.",
description="Override description shown to the LLM when used as a tool.",
json_schema_extra={"rows": 3},
)
query: str = Field(..., min_length=1)
model: Literal["sonar", "sonar-pro", "sonar-reasoning", "sonar-reasoning-pro"] = "sonar"
search_recency_filter: Literal["all", "month", "week", "day", "hour"] = Field(
default="all",
)
return_images: bool = Field(default=False)
return_related_questions: bool = Field(default=False)
model_config = {"extra": "ignore"}
class PerplexitySearchNode(ActionNode):
type = "perplexitySearch"
display_name = "Perplexity Search"
subtitle = "AI Search"
group = ("search", "tool")
description = "AI-powered search using Perplexity Sonar with citations"
component_kind = "square"
tool_name = "perplexity_search"
tool_description = "Search the web using Perplexity Sonar AI. Returns an AI-generated answer with citations and source URLs."
handles = (
{"name": "input-main", "kind": "input", "position": "left", "label": "Input", "role": "main"},
{"name": "output-main", "kind": "output", "position": "right", "label": "Output", "role": "main"},
)
credentials = (PerplexityCredential,)
annotations = {"destructive": False, "readonly": True, "open_world": True}
task_queue = TaskQueue.REST_API
chat_step = "Searched the web"
chat_sources = True
usable_as_tool = True
Params = PerplexitySearchParams
Output = PerplexitySearchOutput
@Operation("search", cost={"service": "perplexity", "action": "sonar_search", "count": 1})
async def search(self, ctx: NodeContext, params: PerplexitySearchParams) -> PerplexitySearchOutput:
body = {
"model": params.model,
"messages": [{"role": "user", "content": params.query}],
}
if params.search_recency_filter and params.search_recency_filter != "all":
body["search_recency_filter"] = params.search_recency_filter
if params.return_images:
body["return_images"] = True
if params.return_related_questions:
body["return_related_questions"] = True
async with ctx.connection("perplexity") as conn:
response = await conn.post(
"https://api.perplexity.ai/chat/completions",
json=body,
headers={"Content-Type": "application/json"},
)
response.raise_for_status()
data = response.json()
choices = data.get("choices") or []
answer = ""
if choices:
answer = (choices[0].get("message") or {}).get("content", "")
citations = data.get("citations") or []
results = [PerplexityResult(url=u) for u in citations]
return PerplexitySearchOutput(
query=params.query,
answer=answer,
citations=citations,
results=results,
images=data.get("images"),
related_questions=data.get("related_questions"),
model=params.model,
)