Search API · Limited beta

Source-aware web search built for AI agents

One API for aggregate, vertical, and strict academic retrieval, safe page and document extraction, bounded crawl, and evidence-oriented research. Built for Codex, Claude Code, and automated workflows.

Apply for beta → View response snapshot

View the Agent Skill on GitHub ↗

Aggregate and academic retrieval
Search-only keys
Evidence provenance
Bounded extraction

api.aifinderkit.com

POST /v1/search balanced

{
  "query": "FastAPI official documentation",
  "limit": 5,
  "mode": "balanced"
}

200 OK sources: 2 cached: false

FastAPI - FastAPI

fastapi.tiangolo.com

bing-rss

FastAPI - FastAPI - tiangolo

fastapi.tiangolo.com/ko

bing-rss

Multi-source Aggregate permitted sources

Provenance Sources retained per result

Isolated Search keys have no model access

Protected Private-network fetches blocked

Capabilities

From finding links to delivering agent-ready evidence

Consistent JSON responses expose sources, degradation warnings, coverage, evidence hashes, and timestamps.

01 POST /search

Aggregate and academic search

Clean snippets, canonicalize URLs, fuse duplicates, and expose an interpretable retrieval score.

fast / balanced / deep
academic-strict / academic-relaxed
DOI, authors, year, and open PDF links

02 POST /batch-search

Batch retrieval

Run up to five independent queries for comparisons, verification, and multi-angle research.

Shared filters and modes
Independent result groups

03 POST /fetch · /map · /crawl

Safe content extraction

Extract HTML, PDF, Office, and GitHub content, or map a bounded same-host static section.

Connection-time public-IP pinning
Byte, page, depth, concurrency, and time limits

04 POST /research

Evidence research bundles

Run root and subqueries, merge cross-query evidence, fetch originals, and report coverage gaps.

Stable citation IDs and content hashes
Explicit degraded and unavailable evidence

Developer Experience

Connect an agent to the web in a few lines

Use familiar Bearer authentication and JSON. The base URL is shared with the model API while search uses isolated exact routes.

Base URL https://api.aifinderkit.com/v1

Works with curl and standard HTTP clients
Ready for Python, Codex, and Claude Code
Programmable errors, sources, and cache state

RESPONSE application/json

{
  "result_count": 3,
  "sources_used": ["searxng", "bing-rss"],
  "warnings": [],
  "cached": false,
  "results": [{
    "title": "FastAPI - FastAPI",
    "url": "https://fastapi.tiangolo.com/",
    "sources": ["bing-rss"],
    "retrieval_score": 0.89,
    "retrieval_signals": {
      "query_relevance": 1.0,
      "source_authority": 0.9
    }
  }]
}

How it works

Queries in. Evidence out.

1
Submit

Choose mode, freshness, and result count.
2
Retrieve

Query permitted sources based on access and availability.
3
Normalize

Canonicalize links, deduplicate, and fuse rankings.
4
Return

Deliver results, sources, warnings, time, and cache state.

Agent Skill

Skip the integration boilerplate

The downloadable skill includes Search, Domains, Batch, Fetch, Map, Crawl, Research, Meta, Doctor, and secure credential setup. Academic presets expose DOI, author, year, and legal open-PDF metadata. No credential is bundled.

v1.3.1 Python 3.10+ Valid key required Zero client dependencies

Extract into your Skills directory
Run configure to store the key securely
Run doctor to verify access

View on GitHub Download ZIP Request a search key

Open source maintained by the AI Finder Kit Organization · ZIP SHA-256 d065fec838eff5dfcea089c56360c36be820bd3a800e5663b832f9c374f583d7 · checksum file

Codex ~/.codex/skills/aifinderkit-search

Claude Code ~/.claude/skills/aifinderkit-search

# Configure a key (none is bundled)
python ~/.codex/skills/aifinderkit-search/\
scripts/aifinderkit_search.py configure

# Verify after installation
python ~/.codex/skills/aifinderkit-search/\
scripts/aifinderkit_search.py doctor

For Claude Code, replace ~/.codex/skills with ~/.claude/skills. Start a new session so the agent discovers the Skill.

Security & transparency

Know where context came from—and where extraction may go

AUTH

Isolated access

Beta applicants receive search-only keys without model permissions.

FETCH

Network boundaries

The service pins a validated public IP at connection time, ignores environment proxies, and rechecks every redirect.

TRACE

Visible provenance

Responses include sources, engines, warnings, evidence IDs, hashes, coverage, and timestamps.

KEYS

Hashed storage

Only irreversible key digests are stored; public skill bundles contain no credentials.

Limited private beta

Apply for Search API access

Priority is given to personal research, multi-device access, and agent workflows. Approved applicants receive a dedicated search-only key.

1Describe your use case
2Manual beta review
3Receive a search-only key by email

Do not submit passwords, existing API keys, or other secrets.

FAQ

Common questions

Can a search key call models?

No. Beta search keys are isolated from model access.

Does it work with Codex and Claude Code?

Yes. Call the JSON API directly or install the downloadable Agent Skill and configure the key through an environment variable.

Does each result expose provenance?

Yes. Results include sources and engines; the response also includes source usage, warnings, cache state, and creation time.

Why is access manually reviewed?

The service is in a limited beta so usage can stay aligned with current capacity and supported workflows.