Skip to content

Commit 8e628a5

Browse files
d-oitd-oit
committed
feat: add Perplexity AI provider support
1 parent b365fd4 commit 8e628a5

File tree

2 files changed

+133
-4
lines changed

2 files changed

+133
-4
lines changed

‎.opencode/agent/perplexity-researcher.md‎

Lines changed: 119 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,119 @@
1+
---
2+
description: >-
3+
Use this agent when the user requests research on a topic that requires
4+
leveraging Perplexity AI for accurate, up-to-date information retrieval and
5+
synthesis, such as querying complex questions, analyzing trends, or gathering
6+
factual data from web sources. This agent utilizes Perplexity's Sonar API,
7+
which integrates real-time web search with natural language processing to
8+
provide responses grounded in current web data with detailed citations. Responses include a 'sources' property containing the websites used for the response.
9+
10+
## Model Selection Criteria
11+
Choose the appropriate Sonar model based on the research task:
12+
- **sonar**: Lightweight and cost-effective for quick factual queries, topic summaries, product comparisons, and current events requiring simple information retrieval.
13+
- **sonar-pro**: Advanced search model for complex queries, follow-ups, and moderate reasoning with grounding.
14+
- **sonar-reasoning**: Fast reasoning model for problem-solving, step-by-step analyses, instruction adherence, and logical synthesis across sources.
15+
- **sonar-reasoning-pro**: Precise reasoning with Chain of Thought (CoT) for high-accuracy tasks needing detailed thinking and recommendations.
16+
- **sonar-deep-research**: Expert-level model for exhaustive research, comprehensive reports, in-depth analyses, and synthesis from multiple sources (e.g., market analyses, literature reviews).
17+
18+
## Prompt Engineering Tips
19+
- Use clear, specific prompts to guide the model; include context, desired format (e.g., summaries, lists), and any constraints.
20+
- For research, request citations, sources, and structured outputs like JSON for better parsing.
21+
- Leverage follow-up prompts for iterative refinement, building on previous responses.
22+
- Specify recency filters or domain restrictions in web_search_options for targeted results.
23+
24+
## Handling Tool Usage and Streaming
25+
All Sonar models support tool usage and streaming. For streaming responses, process chunks incrementally to handle long outputs efficiently. Use streaming for real-time display or to manage large research reports.
26+
27+
## Provider Options Management
28+
- **return_images**: Enable for Tier-2 users to include image responses in results, useful for visual research topics.
29+
- Manage options via providerOptions: { perplexity: { return_images: true } }.
30+
31+
## Metadata Interpretation
32+
- **usage**: Includes citationTokens (tokens used for citations), numSearchQueries (number of searches performed), and cost details.
33+
- **images**: Array of images when return_images is enabled.
34+
- Access via result.providerMetadata.perplexity for monitoring and optimization.
35+
36+
## Proactive Research Strategies
37+
- Schedule periodic queries for ongoing monitoring (e.g., AI ethics developments, market trends).
38+
- Use for trend analysis, competitive intelligence, and automated report generation.
39+
- Combine with tools like Task for multi-step research workflows.
40+
41+
## Recent Advancements
42+
- Introduction of Chain of Thought (CoT) in sonar-reasoning-pro for enhanced reasoning precision.
43+
- Expanded model range including deep research capabilities for exhaustive analyses.
44+
- Improved streaming and tool integration for dynamic, real-time research.
45+
46+
## Actionable Recommendations
47+
- Default to sonar-reasoning-pro for balanced research tasks requiring reasoning and grounding.
48+
- Enable streaming for long-form outputs to improve user experience.
49+
- Monitor metadata for cost optimization and query efficiency.
50+
- Use structured prompts and response formats for better integration with downstream processes.
51+
52+
This agent is ideal for proactive research tasks where direct answers aren't
53+
immediately available and deeper investigation is needed. Examples include:
54+
55+
- <example>
56+
Context: The user is asking about the latest developments in AI safety.
57+
user: "What are the recent advancements in AI safety?"
58+
assistant: "I need to research this using Perplexity AI for comprehensive insights."
59+
<commentary>
60+
Since the user is requesting research on a complex topic, use the Task tool to launch the perplexity-researcher agent to query and synthesize information from Perplexity AI.
61+
</commentary>
62+
assistant: "Now let me use the Task tool to launch the perplexity-researcher agent."
63+
</example>
64+
- <example>
65+
Context: The user is creating an agent for ongoing research monitoring.
66+
user: "Set up a system to monitor AI ethics research."
67+
assistant: "This requires proactive research capabilities."
68+
<commentary>
69+
Since the task involves ongoing research monitoring, use the Task tool to launch the perplexity-researcher agent to periodically query and report on AI ethics developments.
70+
</commentary>
71+
assistant: "I'll use the Task tool to launch the perplexity-researcher agent for this monitoring task."
72+
</example>
73+
- <example>
74+
Context: Using provider options for image responses.
75+
user: "Research the latest in AI image generation and include images."
76+
assistant: "To include images, use provider options with return_images: true (requires Tier-2)."
77+
<commentary>
78+
Launch the perplexity-researcher agent with providerOptions: { perplexity: { return_images: true } } to enable image responses.
79+
</commentary>
80+
assistant: "Launching agent with image options."
81+
</example>
82+
- <example>
83+
Context: Accessing metadata.
84+
user: "What is the usage and sources for this query?"
85+
assistant: "Access providerMetadata.perplexity for usage and images."
86+
<commentary>
87+
After generating text, log result.providerMetadata.perplexity.usage and result.sources.
88+
</commentary>
89+
assistant: "Query executed, metadata available."
90+
</example>
91+
- <example>
92+
Context: Optimizing for deep research with streaming.
93+
user: "Conduct an exhaustive analysis of renewable energy trends."
94+
assistant: "For comprehensive reports, use sonar-deep-research with streaming enabled."
95+
<commentary>
96+
Launch the agent with model: "sonar-deep-research" and stream: true for detailed, real-time output.
97+
</commentary>
98+
assistant: "Initiating deep research with streaming."
99+
</example>
100+
- <example>
101+
Context: Prompt engineering for structured output.
102+
user: "Summarize top AI startups with funding details in JSON format."
103+
assistant: "Use a structured prompt and response_format for JSON output."
104+
<commentary>
105+
Specify response_format: { type: "json_schema", json_schema: { ... } } to get parsed results.
106+
</commentary>
107+
assistant: "Generating structured research summary."
108+
</example>
109+
mode: subagent
110+
model: perplexity/sonar-deep-research
111+
tools:
112+
bash: false
113+
write: false
114+
webfetch: false
115+
edit: false
116+
glob: false
117+
task: false
118+
---
119+

‎opencode.json‎

Lines changed: 14 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -1,16 +1,26 @@
11
{
22
"$schema": "https://opencode.ai/config.json",
3-
"agent": {
4-
5-
},
3+
"agent": {},
64
"provider": {
75
"nvidia": {
86
"models": {
97
"deepseek-ai/deepseek-v3.1-terminus": {
108
"name": "deepseek-v3.1-terminus"
119
}
1210
}
13-
}
11+
},
12+
"perplexity": {
13+
"npm": "@ai-sdk/perplexity",
14+
"name": "Perplexity AI",
15+
"options": {
16+
"baseURL": "https://api.perplexity.ai"
17+
},
18+
"models": {
19+
"sonar-deep-research": {
20+
"name": "Perplexity Sonar Deep Research",
21+
},
22+
}
23+
}
1424
},
1525
"mcp": {
1626
"context7": {

0 commit comments

Comments
 (0)