aiXplain Marketplace Search

aiXplain Marketplace Search is an aiXplain-managed utility tool that searches and inspects the aiXplain marketplace from inside an agent or script. Find agents, models, tools, and integrations by keyword; filter by modality, developer, supplier, or host; pull full details for any asset (function, supplier, pricing, status); and read result counts per asset type.

View the asset on the aiXplain Marketplace.

---

Setup

from aixplain import Aixplain

aix = Aixplain(api_key="<AIXPLAIN_API_KEY>")

---

Get the tool

It is a built-in utility — fetch it by ID.

browse = aix.Tool.get("6960f934f316da19e5f22494")
print(browse.id)

Show output

---

Actions

Action	Input (data)	Returns
search	query (required; "" = wildcard) + optional filters	Counts and top hits per type: {agent, model, tool, integration}, each {items, total}
search_models	query + optional filters	Up to 10 models: [{id, path, name}]
search_tools	query + optional filters	Up to 10 tools
search_agents	query	Up to 10 agents
search_integrations	query	Up to 10 integrations
get_asset_details	assetId	Full metadata for one asset
list_filters	(none)	Valid filter values: developers, hosts, suppliers, categories

Optional filters (arrays, accepted by search and search_models/search_tools): categories (modality/domain), developers, suppliers, hosts, and function (fine-grained task).

note

Only search returns a total per asset type. The search_* actions return at most 10 results and do not paginate, so use search whenever you need a count or a category-wide view.

---

Search across everything

result = browse.run(action="search", data={"query": "translation"})
print(result.data)

Show output

---

Count assets with filters

Pass an empty query as a wildcard and filter by categories to count a whole class of assets. Read the total for the type you care about.

llms = browse.run(action="search", data={"query": "", "categories": ["LLM"]})
print("LLMs:", llms.data["model"]["total"])

openai_llms = browse.run(action="search", data={"query": "", "categories": ["LLM"], "hosts": ["openai"]})
print("OpenAI-hosted LLMs:", openai_llms.data["model"]["total"])

Show output

---

Search one asset type

models = browse.run(action="search_models", data={"query": "whisper", "function": "speech-recognition"})
print(models.data)

Show output

---

Inspect one asset

Use get_asset_details with assetId to read pricing, supplier, function, and status.

details = browse.run(action="get_asset_details", data={"assetId": "66aa869f6eb56342c26057e1"})
print(details.data)

Show output

---

List available filters

filters = browse.run(action="list_filters", data={})
print(list(filters.data.keys()))

Show output

categories includes modality/domain values such as LLM, Speech, Image, Video, OCR, Classification, and Language. Use these values in the categories filter.

---

Use with an Agent

Attach Marketplace Search so an agent can answer catalog questions ("do we have X? how much does it cost? who hosts it? how many LLMs do we have?"). The Marketplace Concierge below answers every question from a live tool result and never invents assets, prices, hosts, or counts.

browse = aix.Tool.get("6960f934f316da19e5f22494")
browse.allowed_actions = [
    "search", "search_models", "search_tools", "search_agents",
    "search_integrations", "get_asset_details", "list_filters",
]

INSTRUCTIONS = """You answer questions about what is available on the aiXplain marketplace, using only the Marketplace Search tool. Never invent assets, prices, hosts, or counts — every fact must come from a tool result.

HOW TO USE THE TOOL (action + data):
- Existence / "do we have X": call `search` with {"query": "<name or keywords>"}. It returns, per asset type (agent, model, tool, integration), a list of items {id, path, name} and a total. If every total is 0, the asset is not available — say so plainly.
- Details / cost / host / supplier / function: take the matching item's id and call `get_asset_details` with {"assetId": "<id>"}. It returns asset_type, function, supplier, hosted_by, developed_by, supports_streaming, pricing {price, unitType}, and status. Report price as "<price> per <unitType>" (e.g. 0.00002 per CHAR). If a field is empty, say it is not listed.
- Counts / "how many X": call `search` with {"query": "", ...filters} and read the `total` for the relevant type. (The search_models / search_tools / search_agents / search_integrations actions cap at 10 results and have no total — use them only to list examples, not to count.)
- Filters (arrays, combine freely on `search`): `categories` for modality/domain (e.g. LLM, Speech, Image, Video, OCR, Classification, Language), `developers`, `suppliers`, `hosts`, and `function` for a fine-grained task. Call `list_filters` (no data) to see valid values if unsure.
- A `search` call REQUIRES a query; pass "" as a wildcard when you only want to filter or count.

HARD RULE — always finish the lookup yourself:
- Whenever the user asks about cost, price, host, supplier, function, or status, you MUST immediately call `get_asset_details` on the best-matching asset's id from your `search` result, in the SAME response. Do NOT stop after `search`, and do NOT ask the user for permission to look up details — just do it and report the answer.
- If `search` returns more than one strong match, fetch details for the most relevant one and mention the other names.

ANSWERING:
- Lead with a direct yes/no or the number, then the supporting details (path, host, supplier, function, price, status).
- When listing matches, show name + path so the user can identify the exact asset.
- Keyword matching is literal — if a multi-word phrase returns nothing, retry with a single distinctive token before concluding the asset is absent.
- Be concise. Use a short bulleted block for details. Do not output raw tool JSON."""

agent = aix.Agent(
    name="Marketplace Concierge",
    description="Answers questions about what is available on the aiXplain marketplace — whether an asset exists, its cost, who hosts it, and how many of a kind exist.",
    instructions=INSTRUCTIONS,
    tools=[browse],
    output_format="markdown",
    max_tokens=4000,
    max_iterations=10,
)
agent.save()

print(agent.run(query="Do we have a Whisper model? Who hosts it and what does it cost?").data.output)

Show output

Parameter	Description
name	Display name for the agent
description	Short summary of what the agent does
instructions	System prompt that constrains the agent to tool-sourced facts and tells it which action to use for existence, detail, and count questions
tools	List of tools the agent can call — here, Marketplace Search with allowed_actions set
output_format	"markdown" so answers render as a tidy yes/no + bulleted detail block
max_tokens	Upper bound on the response length
max_iterations	Maximum tool-call/reasoning steps per run

More questions the Concierge can answer:

print(agent.run(query="How many LLMs are on the marketplace? How many are hosted by OpenAI?").data.output)
print(agent.run(query="Is there a translation model from Google, and what does it cost per character?").data.output)

Show output

---

Notes

• Counts come from search only. The search_* actions cap at 10 results and do not paginate; use search and read total for counts or category-wide views.
• search needs a query. Pass "" as a wildcard when you only want to filter/count.
• Keyword matching is literal. Prefer single distinctive tokens; multi-word phrases can under-match.
• Parameter names differ by action: searches use query; get_asset_details uses assetId.
• It is a built-in utility — there is no connect step and nothing to delete; fetch it with aix.Tool.get(...).