Exa
Also known as: Metaphor Systems, Exa AI
Web search and retrieval API built for AI agents that finds pages by meaning and returns clean contents, with an asynchronous research agent and scheduled monitors.
Exa is a web search and retrieval platform built for AI agents. It runs its own crawler and an independent index of the public web, tracking more than two trillion URLs and serving 100 billion pages, alongside indexes of people, companies, code and scholarly publications.
Exa 3.0, released in October 2026, trained a larger embedding model, made Dynamic Highlights generally available (excerpts chosen across all results, using about 95% fewer tokens than full page text) and added billions of documents, including 1.27 billion ecommerce listings, more than 211 million local businesses, and real time news, sports, weather, market, flight, event, transit and podcast data. Exa returns ranked results with clean, token efficient page contents and highlights for searches written in natural language. It is called from code, through its API, SDKs and a hosted MCP server, and from Claude and ChatGPT through a connector and a plugin, rather than through a consumer search box.
The Search API offers three recommended types: Lite for cost and latency sensitive work, served from Exa's cache, Auto, and Deep, which reasons across several searches and returns structured output with citations. The Contents API extracts text, highlights and summaries from known URLs, including pages rendered with JavaScript and PDFs. Answer returns a cited answer to a question. Monitors run a search on a schedule and post only new findings to a webhook.
Exa Agent is an asynchronous agent for list building, enrichment and deep research that returns JSON validated against the customer's schema with citations for each field, can continue from a previous run, and through Exa Connect can query partner data providers such as Fiber.ai, Similarweb and Baselayer in the same run. Its Ultra effort, added in September 2026, runs for up to three hours within a cost cap the customer sets.
Exa works with OpenAI and Anthropic tool calling, the Vercel AI SDK, OpenRouter, LangChain, LlamaIndex, CrewAI, Pydantic AI and n8n, and offers an endpoint compatible with the OpenAI SDK. Exa says it is SOC 2 Type II certified. Enterprise adds Zero Data Retention, a HIPAA mode, SSO and SCIM for the dashboard, and custom indexes. Exa runs as a hosted service with no self hosted or region option.
Pricing is pay as you go from prepaid credits, with ten dollars of free credit every month plus a ten dollar onboarding bonus. Lite search is four dollars per thousand requests, Auto seven and Deep twelve, contents one dollar per thousand pages, and Agent runs from just over a cent each at the lowest fixed effort.
From our comparison video
Exa vs Firecrawl: One Finds the Pages, One Turns Them Into Data
Our 4 minute video puts Exa side by side with Firecrawl. It covers what each one does and how each is priced, then when to choose which.
Recorded in October 2026. The profile here is live and updates as research is verified.
Vendor details
Canonical URL
https://exa.ai
Category
Agent infrastructure
Subcategory
Web search and retrieval
Funding status
Venture backed. Exa announced a 250 million dollar Series C at a 2.2 billion dollar valuation, led by a16z, on 20 May 2026 (exa.ai/blog/announcing-series-c). Independent.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Exa is API first. It offers a REST API with published OpenAPI specifications, Python and JavaScript SDKs, a hosted MCP server, an endpoint compatible with the OpenAI SDK, and Agent Skills. Integrations include OpenAI and Anthropic tool calling, the Vercel AI SDK and AI Gateway, OpenRouter, Baseten, LangChain, LlamaIndex, CrewAI, Pydantic AI, n8n, Gumloop, Snowflake, Slack and Google Sheets. Exa Connect adds partner data providers inside Agent runs.
In practice
Your RAG pipeline returns keyword matched links that still need cleaning. You call Exa, which retrieves pages by meaning and returns clean, token efficient contents ready for the context window.
You need a list of companies matching open ended criteria, with the people to contact and a source for every field. You start an Exa Agent run with an output schema and get validated JSON back, cited field by field.
You need to track competitor funding rounds without polling the web yourself. You set up an Exa Monitor that runs on a schedule and posts only new findings to your webhook.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
6.5 / 14 capabilities · 46%
| Integrations & Tool Calling | Partial |
|---|---|
|
An agent reads the web and partner data through Exa's tools for search, contents, answer, monitors and the Exa Agent. Exa Connect adds partner sources inside an Agent run, among them Fiber.ai, Similarweb, Baselayer and Polymarket, with Exa handling the provider authentication. Every connection only reads. Exa takes no action in another system and holds no per user credentials for one. Sourceexa.ai/llms.txtread 2026-09-22 |
|
| Workflow Orchestration | Not documented |
|
Inside an Exa Agent run, the agent orchestrates its own tool calls. It fans out searches, splits list building into parallel subtasks and verifies candidates, and that pipeline is Exa's own ready made agent. The customer supplies a query, an output schema, an effort level and data sources, and cannot define, branch or sequence the steps. The Batch API queues many requests to run asynchronously as jobs. There is no workflow runtime where the customer sequences, branches, retries or routes work, and the customer's own logic lives in its own framework. Sourceexa.ai/docs/agent/quickstartread 2026-09-22 |
|
| Knowledge Grounding & RAG | Not documented |
|
The index Exa searches is its own, built from the public web plus select partner data such as people, company and scholarly indexes and premium providers reached through Exa Connect. None of it is the customer's own documents, and Exa names no way for a customer to add its own documents to a structure Exa maintains and queries. Websets imports take a CSV of entities to enrich and filter for a research job. The Enterprise plan includes custom, tailored indexes, and Exa does not say whose content they hold. Sourceexa.ai/docs/search/data/overviewread 2026-09-22 |
|
| Human Oversight & Guardrails | Not documented |
|
The agent reads the web and returns data. It takes no action in another system that a person would need to approve. Job controls let a customer stop an Agent run, keeping the results gathered so far, or cancel it, and pause a scheduled monitor, which stops its automatic runs. Stopping a run ends it, and pausing a monitor halts its schedule, so neither holds work for someone to review. Exa names no approval step, consent checkpoint, escalation rule, guardrail or policy constraint, and no run waits for a person. Sourceexa.ai/docs/monitors/quickstartread 2026-09-22 |
|
| Security, Identity & Governance | Full |
|
The SOC 2 reports and Data Processing Agreement are shared through the Trust Center at trust.exa.ai, and Exa says it is SOC 2 Type II certified. The Enterprise plan carries SSO and SCIM for the Exa Dashboard and Zero Data Retention enabled per team. An Exa organization with SSO and directory sync can let Claude reach Exa's MCP server through Okta, so removing a person in Okta ends their access. Team admins invite members, and a Team Management API creates, rate limits and deletes API keys. HIPAA mode with a BAA is available on Enterprise for eligible cached retrieval. Sourceexa.ai/docs/admin/security/overviewread 2026-09-22 |
|
| Observability & Auditability | Partial |
|
Completed Agent runs are stored with their answer, structured output, citations for each field and a cost breakdown, and can be listed and retrieved later. Each monitor keeps a run history with every run's output and any failure reason. The Team Management API reports usage and cost per API key. While a run streams, its events carry search traces, sources and tool progress grouped by research step. Afterwards only the run's lifecycle events are stored for replay. Outputs and their sources are kept, but Exa does not say whether the step by step trace of prompts, tool calls and retrieved knowledge is kept. It names no audit log, export to monitoring or SIEM systems, or retention period by plan. Sourceexa.ai/docs/agent/quickstartread 2026-09-22 |
|
| Memory & State Persistence | Partial |
|
An Exa Agent run can continue from an earlier one. The previousRunId field carries the earlier run's context into a follow up request, such as asking for ten more results, so the agent picks up where it left off. The team's stored runs can be listed, retrieved and deleted. Continuation is scoped to the team, with no per user scope or editing, and it is switched off under Zero Data Retention. Exa states no retention period and names no memory beyond this continuation. A monitor keeps a record of results it has already returned and uses it to drop duplicates, not as context. Sourceexa.ai/docs/agent/quickstartread 2026-09-22 |
|
| Deployment & Data Residency | Not documented |
|
Delivery is a hosted API. The only regional control blocks requests from sanctioned countries. There is no dedicated or private cloud, self hosted or on premises option, and customers cannot choose a region or pin model traffic or data storage to specific infrastructure. Zero Data Retention sets how long request data is kept. Sourceexa.ai/docs/admin/security/overviewread 2026-09-22 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
The prebuilt piece is Exa Agent, which builds lists, enriches entities and runs deep research, returning cited output validated against the customer's schema. It is also offered in Slack and Google Sheets. Exa publishes Agent Skills for coding agents (Build with Exa, Search and Contents) and example Agent configurations for list building, KYB and job postings. All of them serve one job, research over web and partner data. There is no browsable catalog of workflows, templates or role specific agents beyond the one research agent and three skills. Sourceexa.ai/docs/agent/quickstartread 2026-09-22 |
|
| Triggers & Channel Coverage | Full |
|
Monitors run an Exa search on a schedule the customer sets, from hourly upward. They drop results already returned, synthesize the new findings and post them to the customer's webhook with a signed payload. A monitor can also be triggered on demand, and Websets monitors keep a dataset refreshed on their own schedule. Exa also answers when someone tags @Exa in Slack. It names no way to subscribe to events in other systems. Sourceexa.ai/docs/monitors/quickstartread 2026-09-22 |
|
| Model Flexibility & Routing | Full |
|
Whatever model the customer's agent runs can take Exa's web data. Exa works with OpenAI and Anthropic tool calling, the Vercel AI SDK and AI Gateway, OpenRouter, where it grounds any OpenRouter model, open models on Baseten, LangChain, LlamaIndex, CrewAI and Pydantic AI, and any MCP client can connect. Only Exa's own Agent, Answer and page summaries run on models Exa chooses, and the customer picks the model for its own agent. Sourceexa.ai/docs/llms.txtread 2026-09-22 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Developers call Exa through a REST API with published OpenAPI specifications (search, contents, answer, monitors, batches, the Agent API, and a Team Management API for keys), official Python and JavaScript SDKs, an endpoint compatible with the OpenAI SDK for chat completions and the Responses API, a hosted MCP server with OAuth, and Agent Skills for coding agents. Sourceexa.ai/llms.txtread 2026-09-22 |
|
| Testing, Debugging & Optimization | Not documented |
|
Published evaluations and open source benchmarks measure Exa's own search quality, latency and research performance, not the customer's agent. Exa names no fixtures, test datasets, scoring or quality gates for the customer's agent workflows. Citations and confidence on Exa's output let a builder check sources. Sourceexa.ai/llms.txtread 2026-09-22 |
|
| Browser & Computer Use | Partial |
|
The Contents API extracts clean content from any URL, including pages rendered with JavaScript, PDFs and complex layouts, and fetches live when the indexed copy is older than the caller allows. Exa's crawler reads the public web while respecting robots rules and does not bypass logins, paywalls or CAPTCHAs. It reads and renders pages that have no API through scraping and a headless fetch. Exa names no interactive control of a browser or computer, such as clicking, typing, filling forms or operating a logged in session. Sourceexa.ai/llms.txtread 2026-09-22 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Search and Contents API endpoints from Exa are now served (auto and lite search types) from an EU region in AWS Stockholm (eu-north-1) at api.eu.exa.ai, with the main web index, embeddings, reranking and highlights replicated there. EU teams get region locked API keys that are refused outside the EU endpoint, no fallback to the US, zero data retention by default and an x-exa-processing-region header on every response.
Bears on: Deployment / data residency
View sourcePodcasts are now part of the Exa index: 3.2 million episodes from 119,000 shows, each with a full transcript, across all endpoints. Results include hosts and guests, publish date and a transcript with speaker labels, so an agent can find what a named person said on recent episodes. New episodes appear soon after they air.
Bears on: MCP / tool calling / API
View sourceExa 3.0 brings a larger embedding model, makes Dynamic Highlights generally available at no extra charge, and adds 1.27 billion ecommerce listings, more than 211 million local businesses, real time news, sports, weather and market data, and flights, events, transit and podcasts to its index. The release also adds Exa Lite, a lower cost search type at $4 per 1,000 requests.
Bears on: Pricing / packaging
View sourcePricing
Usage based · $10 free credit/mo plus $10 at sign up · Lite search $4/1k requests
Per 1,000 requests by endpoint and search type, plus per 1,000 pages for contents and summaries; Agent runs per run by effort or per Agent Compute Unit
Included quota
Free Starter plan: $10 credit every month plus a $10 onboarding bonus, access to all endpoints, 10 search queries per second and 50 concurrent Agent runs. The base search price covers up to 10 results.
What is public
Exa publishes a rate card for every endpoint and three plans: a free Starter plan funded with credits, pay as you go, and custom Enterprise.
Billing mechanics
Usage draws down a prepaid credit balance at per-request rates by endpoint and search type; Contents and summaries bill per page and per content type. Agent runs bill a flat price per fixed effort level, or metered compute units plus search calls and contact enrichment, with a per run cost cap the customer can change (Ultra defaults to $20 a run). Exa Connect provider calls bill on top. No seats or minimums; Enterprise adds postpaid invoicing.
Cost watchouts
Results beyond the first 10, each extra content type, AI page summaries, Agent contact enrichment ($0.02 per email, $0.07 per phone) and Exa Connect provider calls each bill on top of the base request.
Variable cost rationale
Pay per request with no minimums keeps small workloads cheap, but cost scales with volume and search type, and deep search or metered Agent runs add up quickly.
Additional watchouts
Search and contents can be separate charges, and /contents bills each content type separately. Deep search and metered Agent runs cost materially more than standard search.
Overage / add-ons
Charged per request from prepaid credits once free credit is used: Lite search $4, Auto $7, Deep $12, Answer $5 and Monitors $15 per 1,000 requests; each result above 10 $1 per 1,000; Contents and AI page summaries $1 per 1,000 pages per content type; Agent $0.012 to $1.00 per fixed effort run, or metered at $0.10 per Agent Compute Unit plus $0.005 per search call. No seats or minimums.
Sales call required
No, self serve available
Free / trial
$10 credit every month plus a $10 onboarding bonus, all endpoints, no payment method
Lowest paid plan
Pay as you go (Developer): Lite search $4/1k requests, Auto $7/1k, Answer $5/1k, Contents $1/1k pages
Commercial notes
Sold to developers and AI teams through free monthly credit, pay as you go billing, a hosted MCP server and framework integrations. Enterprise adds custom indexes, custom rate limits, SLAs and MSAs, Zero Data Retention, SSO and SCIM, HIPAA, a dedicated Slack channel and volume discounts. Startup and education grants offer free credit.
Key ambiguities
Total cost depends on request volume, search type, results and content types pulled, and Agent effort. The free allowance is a monthly dollar credit, not a request count.
Cancellation / refund
Self serve pay as you go with no minimums or seats; usage stops when you stop calling the API. Enterprise terms are contractual.
Support SLA / resale
Standard email support on pay as you go; Enterprise adds a dedicated Slack channel, SLAs, onboarding and invoicing.
Missing data
Enterprise rates, volume discounts and custom index pricing are not published; metered Agent run cost depends on compute units used.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Exa
The closest documented capability profiles to Exa among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Fly.io Sprites6.5 / 14Fuller documented coverage on Integrations & Tool Calling
- Freestyle5.5 / 14A lighter documented profile than Exa
- Morph Cloud5.5 / 14A lighter documented profile than Exa
- Modal8.0 / 14Adds documented Workflow Orchestration and Deployment & Data Residency
- exe.dev7.5 / 14Adds documented Deployment & Data Residency
- Tavily4.5 / 14A lighter documented profile than ExaExa vs Tavily →
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded