LoginSign Up
← Back to all sparks
Perplexity logo

Perplexity

AI-ASSISTANTS
Velocity0.0

Conversational answer engine combining LLMs with live web search.

www.perplexity.ai ↗

Perplexity pivoted from search API to agent platform with February's Agent API GA.

agent-platformmodel-gatewaydeveloper-sdkssearch-apiembeddingsenterprise-distribution
◆Current state
Perplexity now operates as a multi-surface developer platform: Search API, Agent API, Embeddings API, and a growing roster of third-party models routed through one OpenAI-compatible endpoint. The last six months added official Python and TypeScript SDKs, an MCP server, file attachments, and one-click integrations across Cursor, VS Code, Claude Desktop, n8n, and OpenClaw. Security and DX have matured in parallel — one-time-reveal keys, automated rotation, an interactive Search playground.
◆Where it's heading
The center of gravity is shifting from search-augmented chat to autonomous agents that call tools, search the web, and reason over documents. Perplexity is also taking on a model-gateway role, exposing OpenAI, Anthropic, Google, NVIDIA, and xAI models behind its own /v1/agent endpoint. Distribution is moving from direct API integration toward embedded surfaces — MCP, n8n, AWS Marketplace — which lowers procurement and discovery friction for enterprise buyers.
◆Prediction
Expect tighter agent-orchestration primitives next — handoffs, multi-tool plans, persistent memory — and an expansion of the Embeddings API into reranking. The AWS Marketplace listing suggests a deliberate enterprise sales motion is forming around the Agent + Search + Embeddings bundle.

◆Recent moves

  1. 1mo ago

    Agent API: New Third-Party Models

    Another batch of third-party models lands in Agent API — Claude Opus 4.7, GPT-5.5, Grok 4.20 — alongside a one-time-reveal key model, an n8n native node, OpenClaw integration, an AWS Marketplace listing, and a public /v1/models endpoint. The pattern is clear: Perplexity wants to be wherever developers already are, with model choice as a commodity feature rather than a differentiator.

  2. 2mo ago

    Agent API: New Third-Party Models

    Adds GPT-5.4, Nemotron, Claude Sonnet 4.6, and Gemini 3.1 Pro Preview to the Agent API while deprecating older Gemini variants. The endpoint rename from /v1/responses to /v1/agent — with the old path kept as an alias — signals that 'agent' has become the canonical surface, not just a beta entrypoint.

  3. 3mo ago

    Agent API: Now Available

    ⚡ SPARK

    The directional pivot in the timeline: Agent API and Embeddings API both go GA in the same release, repositioning Perplexity from search-and-chat provider to a full-stack agent platform. Everything that follows in March and April — third-party model expansion, AWS Marketplace, n8n — only makes sense as distribution for these two APIs.

  4. 5mo ago

    Model Deprecation: sonar-reasoning Removed

    Routine deprecation of sonar-reasoning is paired with substantive additions: a Media Classifier that decides when to inject images and videos into responses, max_tokens-per-page in Search, a last_updated_filter, and Vercel AI SDK compatibility. The Search API continues to accumulate the granular controls a serious search-data buyer would want.

  5. 6mo ago

    Pro Search: Now Generally Available

    Pro Search exits beta for Sonar Pro with multi-step reasoning, real-time thought streaming, and built-in web_search and fetch_url_content tools — a precursor to the Agent API that landed three months later. One-click MCP server installation across Cursor, VS Code, Claude Desktop, and Claude Code points to where Perplexity sees its distribution leverage.

  6. 7mo ago

    Official Perplexity SDKs

    Python and TypeScript SDKs finally land with type safety, async iterators, and automatic key handling — a DX milestone that signals Perplexity is targeting production builders, not just curious tinkerers. An interactive Search playground and new domain and language filters round out a release focused on developer adoption rather than new capability.