Managing Perplexity API

Perplexity API Management Skill

Manage and analyze Perplexity AI API resources including models, usage, and search-augmented generation.

API Conventions

Authentication

All API calls use Bearer API key, injected automatically.

Base URL

https://api.perplexity.ai

Core Helper Function

#!/bin/bash

perplexity_api() {
    local method="$1"
    local endpoint="$2"
    local data="${3:-}"

    if [ -n "$data" ]; then
        curl -s -X "$method" \
            -H "Authorization: Bearer $PERPLEXITY_API_KEY" \
            -H "Content-Type: application/json" \
            "https://api.perplexity.ai${endpoint}" \
            -d "$data"
    else
        curl -s -X "$method" \
            -H "Authorization: Bearer $PERPLEXITY_API_KEY" \
            "https://api.perplexity.ai${endpoint}"
    fi
}

Output Rules

• Target ≤50 lines per script output
• Use jq to extract only needed fields
• Never dump full API responses

Phase 1: Discovery

#!/bin/bash
echo "=== Available Models ==="
perplexity_api GET "/models" \
    | jq -r '.data[] | "\(.id)\t\(.owned_by // "perplexity")\t\(.context_length // "N/A")"' \
    | column -t | head -15

echo ""
echo "=== API Health Check ==="
RESULT=$(perplexity_api POST "/chat/completions" '{"model": "sonar", "messages": [{"role": "user", "content": "test"}], "max_tokens": 1}')
echo "$RESULT" | jq '{model: .model, usage: .usage, citations: (.citations | length // 0)}' 2>/dev/null \
    || echo "Error: $(echo $RESULT | jq -r '.error.message // "unknown"')"

Phase 2: Analysis

Performance & Rate Limits

#!/bin/bash
echo "=== Model Latency Check ==="
for model in "sonar" "sonar-pro"; do
    START=$(date +%s%N)
    RESULT=$(perplexity_api POST "/chat/completions" "{\"model\": \"$model\", \"messages\": [{\"role\": \"user\", \"content\": \"What is 2+2?\"}], \"max_tokens\": 10}")
    END=$(date +%s%N)
    LATENCY=$(( (END - START) / 1000000 ))
    TOKENS=$(echo "$RESULT" | jq '.usage.total_tokens // 0')
    CITATIONS=$(echo "$RESULT" | jq '.citations | length // 0')
    echo "$model: ${LATENCY}ms, ${TOKENS} tokens, ${CITATIONS} citations"
done

echo ""
echo "=== Rate Limit Status ==="
HEADERS=$(curl -s -D- -o /dev/null -X POST \
    -H "Authorization: Bearer $PERPLEXITY_API_KEY" \
    -H "Content-Type: application/json" \
    "https://api.perplexity.ai/chat/completions" \
    -d '{"model": "sonar", "messages": [{"role": "user", "content": "test"}], "max_tokens": 1}')
echo "$HEADERS" | grep -i "x-ratelimit\|retry-after" | while read line; do
    echo "  $line"
done

Search Quality

#!/bin/bash
echo "=== Search-Augmented Response Check ==="
RESULT=$(perplexity_api POST "/chat/completions" '{"model": "sonar", "messages": [{"role": "user", "content": "Latest news today"}], "max_tokens": 50}')
echo "$RESULT" | jq '{
    model: .model,
    citations_count: (.citations | length),
    citations: .citations[0:5],
    usage: .usage
}'

echo ""
echo "=== Model Comparison ==="
for model in "sonar" "sonar-pro"; do
    RESULT=$(perplexity_api POST "/chat/completions" "{\"model\": \"$model\", \"messages\": [{\"role\": \"user\", \"content\": \"Current weather in NYC\"}], \"max_tokens\": 30}")
    CITATIONS=$(echo "$RESULT" | jq '.citations | length // 0')
    TOKENS=$(echo "$RESULT" | jq '.usage | {input: .prompt_tokens, output: .completion_tokens}')
    echo "$model: $CITATIONS citations, tokens: $TOKENS"
done

Output Format

=== Perplexity API ===
Models Available: <list>

--- Performance ---
<model>: <latency>ms, <tokens> tokens, <citations> citations

--- Rate Limits ---
Requests: <remaining>/<limit>

--- Search Quality ---
Avg Citations: <n> per query

Anti-Hallucination Rules

1. NEVER assume resource names — always discover via CLI/API in Phase 1 before referencing in Phase 2.
2. NEVER fabricate metric names or dimensions — verify against the service documentation or --help output.
3. NEVER mix CLI commands between service versions — confirm which version/API you are targeting.
4. ALWAYS use the discovery → verify → analyze chain — every resource referenced must have been discovered first.
5. ALWAYS handle empty results gracefully — an empty response is valid data, not an error to retry.

Counter-Rationalizations

Common Pitfalls

• Search-augmented: Perplexity models include web search results and citations automatically
• Citations field: Responses include citations array with source URLs
• OpenAI-compatible: API follows OpenAI chat/completions format
• Rate limits: Check x-ratelimit-* response headers for current limits
• Model names: Use sonar or sonar-pro for search-augmented models