balance
ChatGPTThe team's prepaid balance in USD, and any spend currently in flight. Check it when a call is refused for funds, or before a job that will make many calls.
Overview
People and company enrichment, AEO and SEO, ads and social data, plus UGC video and images - all in one place, with no provider API keys to manage.
Tools
call
ChatGPTCall something through treg.
params is the request body for a POST (an object, or an ARRAY of task objects for providers like DataForSEO that expect one) and the query string for a GET. Either a CATALOG endpoint by its id (from catalog_search), or one of THIS TEAM'S own tools as '<tool-name>/<path>' (from my_tools) — e.g. 'render/v1/services' — or a HUB tool by its id '<team-slug>.<name>' (a maker's tool made of tools; params are its inputs, see catalog_get for them). treg injects the credential server-side and relays the provider's response unchanged, so you never hold an API key. Catalog calls on treg's key are metered from the team's prepaid balance; a team's own tool is never metered. Tell the human the price (from catalog_get) before calling anything that costs more than a cent. idempotency_key: pass the SAME key when you are repeating a call whose answer you did not receive — a timeout, a dropped connection, an error on your side after the request went out. treg returns the stored answer, does not call the provider again, and charges nothing the second time; the result carries replayed: true. Use a NEW key (or none) for genuinely new work, even when the parameters are identical: repeating a search to see what changed is a new call, not a retry, and reusing the key would hand you the old answer. Reusing one key for a DIFFERENT request is refused rather than answered. For requests params can't express, the explicit slots mirror the CLI's flags: query is ALWAYS the query string (a list value expands to repeated keys), body is ALWAYS the request body (object/array → JSON; a STRING is sent raw, with content_type naming what it is — sniffed as application/json when it parses as JSON), and headers adds upstream request headers an endpoint needs per-call (e.g. Google Ads' login-customer-id); injected credentials always win over them. Use query + body together for endpoints that split a POST across both (Bright Data's ?dataset_id=… + array body). Giving body implies POST. For multipart requests, put ordinary fields in form and files in uploads as objects with name, filename, content_type and base64 data. Query values are sent the way HTTP spells them: booleans as true/false, nested objects as compact JSON. A null query value is OMITTED from the query string — if an upstream distinguishes an absent parameter from an empty one, send the empty string explicitly rather than null.call_media
ChatGPTCall an audio-producing endpoint through the same treg proxy as
call, returning the successful binary response as native MCP AudioContent. Use this for Fish Audio TTS; use call for JSON endpoints.catalog_get
ChatGPTOne endpoint in full: its parameters, the exact price per call, whether treg serves it on its own key, and the other providers that answer the same capability — with the success rate and speed treg has measured across real calls. Read this BEFORE call() so you can tell the human what it will cost.
catalog_request
ChatGPTThe catalog doesn't have what you need? File a tool request — one sentence saying what capability or provider is missing. Requests are the demand signal that decides which provider gets added next. Use AFTER catalog_search comes up empty, not instead of it.
catalog_search
ChatGPTSearch 3,800+ API endpoints by WHAT YOU WANT TO DO. Use plain task words: 'work email', 'backlinks for a domain', 'tiktok comments', 'keyword search volume'; a provider or platform name lists what it offers. Returns each endpoint's id, provider, price per call, and whether treg can serve it without you owning an API key. Read
verdict: strong = these do it; closest = nearest, check catalog_get; none = not in the catalog, file catalog_request. Use it when a task needs data or an API you have no key for.feedback
ChatGPTProactively share problems and suggestions about treg. Small annoyances that slowed your task down are useful feedback too: confusing results or charges, unclear instructions, unhelpful errors, or missing capabilities you worked around. Report these even if the task succeeded. Describe what you needed and what you observed; you do not need to prove a bug. Categories: quality (tool results), pricing (charges or prices), friction (using treg), other (requests or suggestions). Pass related call IDs in call_ids (CLI: --call-id), not only in message. References are optional. Omit private information, credentials, and raw requests, responses or logs. Report each issue once. Feedback goes to the treg team to improve the product; it does not complete the user's task. Keep going afterward.
my_tools
ChatGPTWhat THIS team has registered and you can call without holding the credential: their own API keys, OAuth connections, vendor CLIs and skills. A team's own tool always wins over treg's catalog key, and those calls are never metered.
resources_list
ChatGPTList durable resources created with treg-provided provider credentials for the current team. Filter Fish voices with provider='fishaudio', kind='voice'.
Capabilities
Writes
Example Prompts
Click any prompt to copy it.
