Tool hosting

Tools grav can offer a model. A tool is a row — one kind executes it, so adding one needs no code change.

3 tools 3 published

This page only reports.

The cards below are the tool definitions as the model receives them — kind, endpoint, auth, arguments, JSON schema, call counters. Editing, enabling/disabling, deleting and test runs live in Admin · + New and /api/tools/<name>.

A tool that never ran here but is manual is normal: grav publishes its schema and the client executes it.

Tools

3 run here

Enabled

all published to models

Calls

0 errors

Kinds installed

one file + one line each

How a model gets these

Ask for the hosted tools on any chat completion, then run the call they lead to.

1 · advertise

curl http://rn.taskd.eu.org/proxy/ai/v1/chat/completions \
  -H "Authorization: Bearer $PROXY_AUTH_KEY" \
  -d '{"model":"<your-model>",
       "messages":[{"role":"user","content":"scrape example.com"}],
       "grav_tools":["firecrawl-scrape"]}'

grav_tools picks which hosted tools to advertise (omit it for all enabled ones). A client that sends its own tools always wins.

2 · execute

curl http://rn.taskd.eu.org/proxy/ai/v1/tools/call \
  -H "Authorization: Bearer $PROXY_AUTH_KEY" \
  -d '{"name":"firecrawl-scrape",
       "arguments":{"url":"https://example.com"}}'

The tool_calls entry of a model answer works verbatim — grav reads function.name / function.arguments too.

`firecrawl-scrape`

HTTP request published no key

0 calls idle

Scrape a single URL and return its content as markdown. A key is optional (keyless tier).

Kind	http One request per call; use {{param}} in the url, headers, query or body.
Endpoint	POST `https://api.firecrawl.dev/v2/scrape`
Auth	bearer
Arguments	url:string * formats:array = ['markdown'] onlyMainContent:boolean = True
JSON schema	show { "additionalProperties": false, "properties": { "formats": { "default": [ "markdown" ], "description": "Output formats, e.g. [\u0027markdown\u0027].", "type": "array" }, "onlyMainContent": { "default": true, "description": "Drop navigation / footer noise.", "type": "boolean" }, "url": { "description": "The absolute URL to scrape.", "type": "string" } }, "required": [ "url" ], "type": "object" }

`firecrawl-search`

HTTP request published no key

0 calls idle

Search the web and return the top results with their content.

Kind	http One request per call; use {{param}} in the url, headers, query or body.
Endpoint	POST `https://api.firecrawl.dev/v2/search`
Auth	bearer
Arguments	query:string * limit:integer = 5
JSON schema	show { "additionalProperties": false, "properties": { "limit": { "default": 5, "description": "How many results to return.", "type": "integer" }, "query": { "description": "The search query.", "type": "string" } }, "required": [ "query" ], "type": "object" }

`http-get`

HTTP request published no key

0 calls idle

Fetch any URL over GET and return the raw body as text.

Kind	http One request per call; use {{param}} in the url, headers, query or body.
Endpoint	GET `{{url}}`
Auth	none
Arguments	url:string *
JSON schema	show { "additionalProperties": false, "properties": { "url": { "description": "Absolute http(s) URL to fetch.", "type": "string" } }, "required": [ "url" ], "type": "object" }