菜单
Exa AI search API via CLI. Use when the user wants to search the web, find code examples, extract page content, or get AI answers with sources.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Hand off the current task to a fresh background agent running in its own workmux worktree, passing a self-contained prompt synthesized from this conversation's context. Use when the user wants to outsource, delegate, or "hand off" implementation work to a separate agent session (e.g. "handoff this fix", "outsource it to a ccd agent", "spin up an agent to do X").
Delegate a code review to pi (`pi /skill:review`) running in a tmux window, then dispatch subagents to fix each finding and check it off the review file. Use when the user wants an automated review-and-fix loop over their current changes.
Teach a topic using the Feynman technique — explain simply, have the user re-explain, identify gaps, re-teach, and loop until they can explain it on their own. Use when the user wants to deeply understand a topic, test their understanding, or mentions "Feynman".
Build a focused 20-hour learning plan for a topic, structured as 10 two-hour sessions following the Pareto principle (20% that drives 80% of results). Use when the user wants to learn a topic fast, build a study plan, or create a structured curriculum.
Create or update a GitHub pull request for the current branch, committing pending changes first. Use when the user wants to open a PR, update an existing PR, or ship a branch for review.
Ask questions about specific GitHub code repositories using DeepWiki. Use when the user asks about a repo's code, architecture, configuration, modules, APIs, examples, or behavior and provides a GitHub repository in owner/repo form or as a GitHub URL.
| name | exacli |
| description | Exa AI search API via CLI. Use when the user wants to search the web, find code examples, extract page content, or get AI answers with sources. |
--toon when fetching content, --json only when piping to jqexacli login to store in OS keychain--text to include full content, --highlights for snippets--poll with research tasks to wait for resultsPick the right command first — wrong command = wrong output:
| Goal | Command |
|---|---|
| Find code examples / implementations | exacli code |
| Web search with content | exacli search --text or --highlights |
| Get a synthesized answer with citations | exacli answer |
| Deep multi-step research report | exacli research --poll (fall back to exacli answer on 404) |
| Fetch specific URL content | exacli contents <url> |
| Find pages similar to a URL | exacli similar <url> |
Measured on 10 results with no content flags (--text/--highlights/--summary off):
| Flag | Chars | Est. tokens | Best for |
|---|---|---|---|
| (none) | ~3,300 | ~830 | Agent contexts with metadata only — omits null/empty fields |
--toon | ~4,200 | ~1,050 | Agent contexts with content — flat key: value, no formatting overhead |
--json | ~5,000 | ~1,250 | Scripting only — most verbose due to null fields + structural chars |
Key insight: --json is ~50% larger than the default for the same data because it includes "highlights": null, "highlightScores": null, "text": "", "summary": "" on every result. Prefer default or --toon to keep context small; reach for --json only when piping to jq.
# Metadata-only: default is most compact
exacli search "query"
# With content: --toon is most compact
exacli search "query" --text --toon
# Scripting: --json when you need jq field extraction
exacli search "query" --json | jq '.results[] | {title, url}'
exacli research-status "task-id" --json | jq '{status, output}'
Search for code examples using the Exa Code API. Use this when the user needs actual code, implementations, or code-focused results — not search.
exacli code <query> [options]
| Option | Description |
|---|---|
--tokens-num <n> | Token budget: dynamic, 1000, 5000, 50000 (default: dynamic) |
exacli search "query" [options]
| Option | Description |
|---|---|
--num-results <n> | Number of results (default: 10) |
--type <auto|fast|deep|instant> | Search type |
--text | Include full text content |
--highlights | Include relevant highlights |
--summary | Include AI-generated summary |
--category <category> | Filter by category |
--include-domains <list> | Comma-separated domains to include |
--exclude-domains <list> | Comma-separated domains to exclude |
--start-date <date> | Start date (ISO format) |
--end-date <date> | End date (ISO format) |
--autoprompt | Use autoprompt to enhance query |
Retrieve content from specific URLs.
exacli contents <url...> [options]
Options: --text, --highlights, --summary, --max-age-hours <n>
Find pages similar to a given URL.
exacli similar <url> [options]
Options: --num-results <n>, --exclude-source-domain, --text, --highlights, --summary, --category <category>
Get AI-powered answers with source citations.
exacli answer "query" [options]
Options: --text, --model <exa|exa-pro>, --stream, --system-prompt <text>
Create automated research tasks. If this returns a 404, fall back to exacli answer.
exacli research "instructions" [options]
Options: --model <fast|regular|pro>, --poll, --poll-interval <ms>, --timeout <ms>
exacli research-status <task-id>
exacli research-list [--limit <n>] [--cursor <token>]
exacli login # store API key in OS keychain
exacli logout # remove API key from OS keychain
| Flag | Description |
|---|---|
--api-key <key> | Exa API key |
--json | Output raw JSON |
--toon | Output compact TOON format |
-h, --help | Show help |
company, research paper, news, pdf, tweet, personal site, financial report, people
| Type | Description |
|---|---|
auto | Automatically chosen (default) |
fast | Quick results |
deep | Thorough, best for research |
instant | Lowest latency |
| Model | Description |
|---|---|
fast | Quick facts |
regular | Balanced (default) |
pro | Highest quality |
# Code search (use code, not search, for code examples)
exacli code "rate limiter implementation Go" --tokens-num 5000
# Semantic search with content
exacli search "AI startups" --num-results 5 --text --toon
# Deep search for research
exacli search "transformer architecture" --type deep --highlights --toon
# Filter by category and date
exacli search "startup funding" --category news --start-date 2024-01-01
# Domain-specific search
exacli search "research" --include-domains "arxiv.org,openai.com"
# Extract content from URLs
exacli contents "https://example.com/article" --text --toon
# Find similar pages
exacli similar "https://openai.com/research" --exclude-source-domain
# AI-powered answers
exacli answer "What is quantum computing?" --stream
# Research with polling (falls back to answer on 404)
exacli research "Latest AI developments" --poll
# fallback:
exacli answer "Latest AI developments" --text
# Check research status
exacli research-status "task-id" --json | jq '{status, output}'
# List research tasks
exacli research-list --limit 10 --json | jq '.data[] | {id, status}'
| Deep multi-step research report | exacli research --poll (fall back to exacli answer on 404) |
| Fetch specific URL content | exacli contents <url> |
| Find pages similar to a URL | exacli similar <url> |
Measured on 10 results with no content flags (--text/--highlights/--summary off):
| Flag | Chars | Est. tokens | Best for |
|---|---|---|---|
| (none) | ~3,300 | ~830 | Agent contexts with metadata only — omits null/empty fields |
--toon | ~4,200 | ~1,050 | Agent contexts with content — flat key: value, no formatting overhead |
--json | ~5,000 | ~1,250 | Scripting only — most verbose due to null fields + structural chars |
Key insight: --json is ~50% larger than the default for the same data because it includes "highlights": null, "highlightScores": null, "text": "", "summary": "" on every result. Prefer default or --toon to keep context small; reach for --json only when piping to jq.
# Metadata-only: default is most compact
exacli search "query"
# With content: --toon is most compact
exacli search "query" --text --toon
# Scripting: --json when you need jq field extraction
exacli search "query" --json | jq '.results[] | {title, url}'
exacli research-status "task-id" --json | jq '{status, output}'
Search for code examples using the Exa Code API. Use this when the user needs actual code, implementations, or code-focused results — not search.
exacli code <query> [options]
| Option | Description |
|---|---|
--tokens-num <n> | Token budget: dynamic, 1000, 5000, 50000 (default: dynamic) |
exacli search "query" [options]
| Option | Description |
|---|---|
--num-results <n> | Number of results (default: 10) |
--type <auto|fast|deep|instant> | Search type |
--text | Include full text content |
--highlights | Include relevant highlights |
--summary | Include AI-generated summary |
--category <category> | Filter by category |
--include-domains <list> | Comma-separated domains to include |
--exclude-domains <list> | Comma-separated domains to exclude |
--start-date <date> | Start date (ISO format) |
--end-date <date> | End date (ISO format) |
--autoprompt | Use autoprompt to enhance query |
Retrieve content from specific URLs.
exacli contents <url...> [options]
Options: --text, --highlights, --summary, --max-age-hours <n>
Find pages similar to a given URL.
exacli similar <url> [options]
Options: --num-results <n>, --exclude-source-domain, --text, --highlights, --summary, --category <category>
Get AI-powered answers with source citations.
exacli answer "query" [options]
Options: --text, --model <exa|exa-pro>, --stream, --system-prompt <text>
Create automated research tasks. If this returns a 404, fall back to exacli answer.
exacli research "instructions" [options]
Options: --model <fast|regular|pro>, --poll, --poll-interval <ms>, --timeout <ms>
exacli research-status <task-id>
exacli research-list [--limit <n>] [--cursor <token>]
exacli login # store API key in OS keychain
exacli logout # remove API key from OS keychain
| Flag | Description |
|---|---|
--api-key <key> | Exa API key |
--json | Output raw JSON |
--toon | Output compact TOON format |
-h, --help | Show help |
company, research paper, news, pdf, tweet, personal site, financial report, people
| Type | Description |
|---|---|
auto | Automatically chosen (default) |
fast | Quick results |
deep | Thorough, best for research |
instant | Lowest latency |
| Model | Description |
|---|---|
fast | Quick facts |
regular | Balanced (default) |
pro | Highest quality |
# Code search (use code, not search, for code examples)
exacli code "rate limiter implementation Go" --tokens-num 5000
# Semantic search with content
exacli search "AI startups" --num-results 5 --text --toon
# Deep search for research
exacli search "transformer architecture" --type deep --highlights --toon
# Filter by category and date
exacli search "startup funding" --category news --start-date 2024-01-01
# Domain-specific search
exacli search "research" --include-domains "arxiv.org,openai.com"
# Extract content from URLs
exacli contents "https://example.com/article" --text --toon
# Find similar pages
exacli similar "https://openai.com/research" --exclude-source-domain
# AI-powered answers
exacli answer "What is quantum computing?" --stream
# Research with polling (falls back to answer on 404)
exacli research "Latest AI developments" --poll
# fallback:
exacli answer "Latest AI developments" --text
# Check research status
exacli research-status "task-id" --json | jq '{status, output}'
# List research tasks
exacli research-list --limit 10 --json | jq '.data[] | {id, status}'