Perplexity

Perplexity (Sonar API)

Perplexity's Sonar models search the web, synthesize an answer, and return citations in one HTTP call. This skill talks to the official API at https://api.perplexity.ai directly with curl — no CLI to install. Every recipe shows a jq command and a python3 -c fallback, because jq may be absent.

Auth

The key lives in PERPLEXITY_API_KEY. Process environment always wins; the shared env file only fills it when unset.

bash / zsh:

[ -f ~/.config/elnora-websearch/.env ] && while IFS='=' read -r k v; do [ -n "$k" ] && [ -z "$(eval echo \$$k)" ] && export "$k=$v"; done < ~/.config/elnora-websearch/.env

PowerShell:

Get-Content "$env:USERPROFILE\.config\elnora-websearch\.env" -ErrorAction SilentlyContinue | ForEach-Object {
  $k,$v = $_ -split '=',2
  if ($k -and -not (Get-Item "env:$k" -ErrorAction SilentlyContinue)) { Set-Item "env:$k" $v }
}

No key yet: create one at https://console.perplexity.ai (create an API Group first, then the API Keys tab — the key is shown only once). Billing is prepaid credits; a card is required at setup but nothing is charged until you buy credits. New accounts are Tier 0.

Router — which recipe

User intent	Recipe	Model	Reference
Find URLs / sources / recent news, raw ranked results	search	(Search API, no LLM)	references/search.md
Quick synthesized answer with citations	search	sonar	references/search.md
General Q&A, pick depth vs speed	ask	sonar / sonar-pro	references/ask.md
Compare, analyze, evaluate tradeoffs (chain-of-thought)	reason	sonar-reasoning-pro	references/reason.md
Literature review, market/competitive research, due diligence	research	sonar-deep-research	references/research.md
HTTP errors, rate limits, cost	—	—	references/errors-and-limits.md

Start cheap (sonar), escalate only when the answer lacks depth. Deep research is expensive and slow — read references/research.md before running it.

Models

Model	Speed	RPM (Tier 0)	Best for
sonar	2–5s	50	Quick facts, URLs, current events
sonar-pro	5–15s	50	Complex queries, more citations, multi-step search
sonar-reasoning-pro	10–30s	50	Comparisons, technical analysis (emits chain-of-thought)
sonar-deep-research	30–120s+	5	Literature reviews, comprehensive reports

Approximate token pricing (verify at https://docs.perplexity.ai/getting-started/pricing): sonar ~$1/$1 per 1M in/out; sonar-pro ~$3/$15; sonar-reasoning-pro ~$2/$8; sonar-deep-research $2/$8 plus per-search and per-citation-token charges. All Sonar calls add a per-request search fee ($5 per 1,000 searches); sonar-pro and deep research may run several searches per call.

Endpoints

• Chat completions (all Sonar models): POST https://api.perplexity.ai/chat/completions — OpenAI-compatible body.
• Search API (raw ranked results, no LLM synthesis): POST https://api.perplexity.ai/search.

The one call to remember

curl -sS https://api.perplexity.ai/chat/completions \
  -H "Authorization: Bearer $PERPLEXITY_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"model":"sonar","messages":[{"role":"user","content":"What is quantum error correction?"}]}'

Extract the answer and its sources:

# with jq
RESP=$(curl -sS https://api.perplexity.ai/chat/completions \
  -H "Authorization: Bearer $PERPLEXITY_API_KEY" -H 'Content-Type: application/json' \
  -d '{"model":"sonar","messages":[{"role":"user","content":"What is quantum error correction?"}]}')
echo "$RESP" | jq -r '.choices[0].message.content'
echo "$RESP" | jq -r '.search_results[]? | "\(.title) — \(.url)"'

# without jq
echo "$RESP" | python3 -c 'import sys,json;d=json.load(sys.stdin);print(d["choices"][0]["message"]["content"])'

Response shape (fields the recipes rely on):

{
  "id": "...",
  "model": "sonar",
  "choices": [{"index":0,"finish_reason":"stop","message":{"role":"assistant","content":"answer text"}}],
  "citations": ["https://source1.com", "https://source2.com"],
  "search_results": [{"title":"...","url":"https://...","snippet":"...","date":"2026-01-01"}],
  "usage": {"prompt_tokens":50,"completion_tokens":200,"total_tokens":250}
}

citations is a flat URL list; search_results carries the same URLs with title/snippet/date. Prefer search_results when you need titles. Both may be empty when the model answers without searching.

Cross-platform note

All curl recipes run as-is in bash/zsh and in PowerShell 7+ (curl is not aliased there when curl.exe is on PATH; if curl resolves to Invoke-WebRequest, call curl.exe). On Windows PowerShell 5.1, use curl.exe explicitly and single-quote the JSON with escaped inner quotes, or pipe the body from a file.