## What treg is **Treg** (at [treg.to](https://treg.to)) is an open-source "tool gateway" for AI agents — its own tagline is **"OpenRouter for agent tools"**. Instead of an agent picking which LLM to call through one unified API and key (like OpenRouter does for models), treg lets an agent pick which third-party *tool/API* to call — SEO data, social listening, lead enrichment, ad intelligence, scraping, and more — through **one base URL, one bearer token, and pay-per-call pricing**, instead of you provisioning and paying for 40+ separate SaaS subscriptions and OAuth apps. The `treg.to/app` path you referenced isn't a distinct product — it lands you in the same hosted dashboard/CLI ecosystem at treg.to where sign-in, catalog browsing, and account management happen.[1][2][3][4] It's built by Superdesign (Jason Zhou), launched publicly on Product Hunt in September 2026, and is open source under Apache 2.0 (self-hostable).[2][5] ## What it does - **Catalog of 2,600–3,270+ endpoints across ~60–82 providers**, spanning SEO/AEO (keyword volume, rank tracking, backlinks — Semrush/Moz/Serpstat/Majestic-style data), social listening (TikTok, Reddit, YouTube, Instagram, X), people/company enrichment (work email finding à la Hunter), advertising intelligence (Meta/Google/TikTok/LinkedIn ad libraries), e-commerce, app-store reviews, and web scraping.[1][6][7] - **Task-based discovery**: your agent searches for what it wants to *do* ("backlinks for a domain," "find a work email") rather than needing to know which vendor sells it. Treg shows competing providers side by side with measured success rate, speed, and price so the agent picks based on evidence.[1][3] - **Credential-injecting proxy**: the agent never holds provider API keys. Treg stores secrets server-side and injects them into the real upstream request, relaying the response verbatim — so a provider's API changing doesn't break your integration, and no secret ever touches your machine.[3][5] - **Pay-per-call billing at 0% markup**, prepaid balance, with new teams starting with $1.00 free credit — versus committing to a $50–$139/mo subscription for tools you'll use occasionally.[1][4] - **Bring-your-own-key (BYOK)**: if your team already pays for a provider (Semrush, Apollo, etc.), connect that key and those calls bypass metering entirely — your own key always wins over treg's.[3][5] ## How it works technically The credential ladder for any catalogued call resolves in order: (1) your team's own registered tool/key for that provider, (2) a stored secret injected via a "virtual tool," (3) a verified public route needing no key (served free), or (4) treg's own key billed against your prepaid balance. The proxy's core rule is that it **relays, never models** the upstream API — it doesn't wrap or reinterpret responses, it just forwards the real request with credentials injected, which is why upstream API changes don't break it.[3][5] Installation is a single shell command that also installs the CLI, plus native support for MCP and Claude Code: ``` curl -fsSL https://treg.to/install.sh | sh treg login treg catalog search "backlinks for a domain" treg call --query ... treg balance ``` For self-hosting: `curl -fsSL https://treg.to/selfhost.sh | sh`, or run from source with Python/uv and a Postgres/SQLite database plus a durable Fernet encryption key [8][5]. For Claude specifically, it's also installable as a plugin (`/plugin marketplace add superdesigndev/treg`) or via the Claude Connectors Directory at `treg.to/mcp/v2/`, and any coding agent can be onboarded in one shot by pointing it at `treg.to/llms.txt`, which teaches the whole protocol [5][3]. ## User sentiment Sentiment is early but positive, with some pointed technical questions still unanswered publicly: - On Product Hunt (4.0/1 formal review, 245 followers as of launch), commenters called it "exactly the kind of infrastructure AI agents need," with one user reporting they spent under $1 and got useful TikTok/Instagram analytics plus a "solid best practices breakdown" within minutes of setup.[2] - A comparison to OpenRouter's cost-transparency UX drew praise — one commenter specifically liked seeing granular per-call cost the same way OpenRouter shows per-token LLM costs.[2] - Skeptical/unresolved questions from the community: how treg handles **GDPR/EU AI Act compliance**, and — more pointedly — since the proxy sees full request/response bodies (not just auth tokens) for things like leads, ads, and scraping data, what the **logging/retention policy** is for that traffic (true zero-log passthrough vs. temporary retention for billing-dispute debugging). Treg had not publicly answered this at time of the thread.[2] - Third-party coverage (toolworthy.ai, selfhosttools.com) frames it favorably as solid infrastructure but flags it as **"early infrastructure"** and **"sensitive control-plane software"** — meaning it should be treated carefully since it centralizes many credentials, not as a casual convenience tool.[8][9] - On GitHub it has meaningful open-source traction: 1,400 stars, 144 forks, 1,700+ commits, and very active daily development (commits through Sept 15, 2026).[5] ## Use cases for you and full/REFIT Given your REFIT work involves SEO/content pipelines, lead enrichment (you've used Apollo via n8n), and AI-automation client work, treg maps well onto several existing pain points: | REFIT workflow | How treg helps | |---|---| | Lead procurement (replacing/supplementing Apollo) | Verified work-email finding and company enrichment across 11+ providers at fractions of a cent per verified record, with misses free — avoids the Waterfall-enrichment overage incident you hit before [4] | | SEO/content pipeline research | Keyword volume, rank tracking, backlink audits, and AI-visibility (GEO) data without paying full Semrush/Moz/Majestic subscriptions just to pull occasional data for content pieces [1][6] | | Competitor/market research for cohort or client audits | Pull live Meta/Google/TikTok/LinkedIn ad-library data and social listening (Reddit, YouTube, TikTok) to inform content strategy or client competitive audits [4] | | n8n automation nodes | Since treg is just HTTP with a bearer token, it drops into n8n as a generic HTTP Request node — one credential covers dozens of downstream data sources instead of managing separate API keys per provider | | Client Rescue Sprint engagements | For the $2,000 "fix one broken automation" offer, treg's BYOK model means you can plug into a client's existing SEO/enrichment subscriptions rather than requiring them to buy new tool accounts | | Self-hosting for data control | Since it's AGPL/Apache-2.0 open source, you could self-host the registry on your own Hostinger/n8n-adjacent infra to keep full control over credential storage rather than trusting the hosted treg.to relay with sensitive lead/client data — relevant given the unresolved logging-retention question above [8][5] | The main caution for REFIT specifically: since the Product Hunt thread flagged unclear data-retention/GDPR posture on the hosted service, if you route real client or prospect data (emails, ad data, enrichment) through it, self-hosting the open-source registry is the safer default until treg publishes a clearer retention policy.[2][8] What treg is treg combines five things: 1. A catalog of external tools    - SEO and keyword research    - SERP and ranking data    - Backlink and domain analysis    - AI search visibility    - Social posts, profiles, comments, and trends    - People and company enrichment    - Email discovery and verification    - Advertising intelligence    - Google Ads and Meta Ads data    - YouTube, TikTok, Instagram, Reddit, LinkedIn, and X data    - Image and video generation    - Market data    - Slack, Telegram, and GitHub tools 2. A credential proxy    - The agent calls treg.    - treg selects the relevant registered tool or catalog endpoint.    - treg injects the credential server-side.    - The upstream response is relayed back. 3. A private team registry    - A team can register its own APIs.    - It can store secrets, OAuth connections, command-line tools, and agent skills.    - Team members can use those resources without seeing the underlying secrets. 4. A command-line interface    - Search the catalog.    - Call APIs.    - Run registered CLIs.    - Upload local .env credentials and skills.    - Manage teams, agents, budgets, tools, and audit records. 5. An MCP and agent integration layer    - The service exposes MCP access.    - It provides an agent-facing llms.txt file and a skill.md file.    - It supports Claude Code, Cursor, Codex, OpenClaw, and other agent environments. The public app at treg.to/app currently redirects to the public landing page. Clicking Start free opens a sign-in dialog with GitHub, Google, or emailed sign-in code options. --- How it works 1. Search by capability The intended workflow is to ask for the task rather than name a vendor. For example: Find the monthly search volume and CPC for these keywords or: Find verified work emails for people matching this role and company type You can search the catalog directly: treg catalog search "backlinks for a domain" treg catalog get hunter.people.email.find The catalog shows the endpoint parameters, price, provider, measured success rate, and speed where available. 2. Call the endpoint through one base URL A catalog call looks like this: treg call hunter.people.email.find \ --query domain=example.com \ --query full_name="Alex Example" At the raw HTTP level, the same idea uses: curl "https://treg.to/call/" \ -H "X-Treg-Token: $TREG_TOKEN" For a tool your team owns, treg can proxy the real upstream URL: https://treg.to/call/https://api.example.com/v1/resource The method, path, query string, body, and most headers are passed through. treg adds the appropriate credential server-side. 3. Choose between providers For one capability, treg may list several providers. For example, backlink data may be available from: - Moz - Majestic - DataForSEO - Semrush - Serpstat - SE Ranking The catalog compares: - Price - Input requirements - Observed success rate - Speed - Last successful check[7:01 AM]treg generally does not silently choose or fail over between providers. The agent or user chooses. Routed capability endpoints are the exception. Those can select a provider and report which provider served the request. 4. Pay per call The pricing model is designed for occasional agent use. Instead of paying for a full seat in a service such as Semrush, Moz, Apollo, Hunter, or Crunchbase, treg charges per call or per result at the provider's rate. The site claims: - No additional markup - Price shown before a call - A one-time $1.00 signup credit for eligible verified teams - Your own provider key takes priority - Calls using your own key are generally unmetered by treg The cost model needs careful testing because “per call” may actually mean per result, per target, per keyword, or per generated asset. The catalog states the unit, but the calling agent still needs to report the actual response cost. 5. Audit and control usage The system includes: - Call IDs - Usage records - Balance tracking - Idempotency keys for safe retries - Team budgets - Per-tag usage attribution - Per-agent tokens - Role-based organization access - Tool access restrictions - Secret health checks - Credential revocation - Call replay behavior for certain failed retries This is one of the more serious parts of the product. It is trying to solve the problem of agents making calls that are difficult to attribute, retry, or bill. --- What is in the catalog The live public pages are changing quickly, and their counts currently disagree: - Landing page: 2,630 endpoints and 47 providers - API docs: 2,630 catalogued provider endpoints - Catalog page: 2,800+ tools and 84 platforms - llms.txt and skill.md: 3,300+ endpoints and 77 providers - GitHub marketing material: references both older and newer counts That inconsistency is important. The catalog itself should be treated as the current source of truth. The main capability groups are: SEO and AEO - Keyword volume - Keyword ideas - CPC and competition - SERP results - Domain ranking keywords - Rank tracking - Backlink profiles - Referring domains - Domain authority - AI Overview citations - Brand mentions in AI answers - Google Search Console - Google Analytics - Google Business Profile Social and content intelligence - TikTok profiles and posts - Instagram posts and reels - YouTube videos, channels, comments, and transcripts - Reddit posts and comments - X posts and profiles - LinkedIn posts and pages - Creator discovery - Engagement sampling - Trend and sound discovery People and company data - Work email discovery - Email verification - Phone lookup - LinkedIn profile data - Person enrichment - Company enrichment - Firmographics - Funding and investor data - Company news - Hiring signals - Contact search Advertising - Meta Ad Library - Google Ads Transparency - Competitor PPC keywords - Google Ads campaign data - Meta Ads campaign data - TikTok Ads - Pinterest Ads - Snapchat Ads - Microsoft Ads - Creative research Media generation - Image generation - Video generation - Seedance - Veo - Wan - Gemini Image - GPT Image - Seedream - MiniMax and related routes Team and developer tools - Slack - Telegram - GitHub - Stripe - Vendor CLIs - Internal HTTP APIs - OAuth connections - Skills and bundles --- How installation works There are several paths. Option 1: Agent-first setup The simplest advertised path is to point an agent at: https://treg.to/llms.txt The file explains the registry, the catalog, authentication, calling conventions, pricing, retries, team management, and feedback system. The official site also provides: https://treg.to/skill.md That is intended to be saved as an agent skill. Option 2: CLI installation The official quickstart is: curl -fsSL https://treg.to/install.sh | sh treg login Login options include: treg login treg login --email you@company.com treg login --token [7:01 AM]The normal login routes are: - GitHub - Google - Email one-time code - Organization token for agents and CI Then create or join a team and search the catalog: treg catalog search "backlinks for a domain" treg catalog get hunter.people.email.find treg balance I would inspect the installer before running it on a production machine. The official command is convenient, but it is still a remote shell installer. Option 3: Claude Code plugin The GitHub repository documents a Claude Code plugin route: /plugin marketplace add superdesigndev/treg /plugin install treg@treg The repository also references: npx skills add superdesigndev/treg -s treg The -s treg matters because the repository contains internal development skills as well as the user-facing treg skill. Option 4: MCP The public documentation lists: https://treg.to/mcp The exact MCP installation method depends on the client. The GitHub documentation describes a first-run flow that installs the CLI, signs in, and then installs treg MCP access. For Hermes, this would need a deliberate connector decision. I would not add it globally without first determining: - Which Hermes profile receives the configuration - Whether the token is stored securely - Whether the MCP server receives the full catalog or only selected capabilities - Whether the agent can spend balance without an approval gate - Whether the connection is team-scoped --- User sentiment Verdict Sentiment is positive toward the problem and the concept. Evidence for sustained product satisfaction is still limited.​ The last-30-days research covered August 17 through September 16, 2026. It found: - 10 X posts - 11 YouTube items - 9 Reddit items, with Reddit partially rate-limited - 3 GitHub items - 7 web pages - A large amount of irrelevant TikTok, Pinterest, LinkedIn, and YouTube material caused by the broad agent-tool topic The strongest directly relevant signals were positive: - An X post with 83 likes and 79 replies praised treg for avoiding lock-in to one data provider. - Another post described using multiple vendors to improve the chance of finding a valid work email at an acceptable cost. - A founder announcement received roughly 220 likes. - Several explanatory videos framed treg as a practical response to the problem of agents needing many paid APIs. - The GitHub repository is active and has continued receiving frequent changes. The appeal is clear: 1. Avoid buying a full subscription for a tool used twice a month. 2. Avoid wiring every agent to a different vendor API. 3. Keep provider credentials off the agent machine. 4. Compare providers by price and observed performance. 5. Give an agent one stable interface while providers change behind it. The caution is equally clear: - There are few mature independent product reviews. - Much of the public discussion is promotional or explanatory. - Several retrieved videos were unrelated false positives. - Reddit coverage was incomplete because of rate limiting. - Public inventory counts do not agree. - The research shows curiosity and adoption interest more strongly than it shows long-term reliability. So I would describe the sentiment as early positive interest, not proven trust. --- How this could help full/REFIT 1. SEO and AEO research This is the clearest fit. An agent could: - Pull keyword volume and related queries. - Check how full/REFIT pages appear in search. - Compare competitor ranking keywords. - Inspect backlinks and referring domains. - Track AI Overview or answer-engine citations. - Run recurring brand visibility checks. - Pull Search Console and Analytics data through Paul's own connected accounts. This could support the Revenue Content Engine without requiring separate seats for every SEO provider. 2. Content demand research treg could give an agent access to: - TikTok trends - YouTube transcripts and statistics - Reddit discussions - X posts - Instagram content[7:01 AM]- Creator discovery - Comment mining - Competitor ad creative For full/REFIT, the practical workflow would be: Find recent public conversations about a specific operational AI problem. Group the recurring pain points. Identify the language people actually use. Separate high-intent problems from general AI enthusiasm. Return source URLs and dated evidence. That aligns with the requirement for net-new public problem statements. It does not require relying on Paul's existing contacts. 3. Public-signal prospect research treg could support research into strangers who have publicly posted about a relevant problem. For example: - Find companies discussing broken AI workflows. - Identify public hiring or technology signals. - Find public posts that indicate a need for workflow reliability. - Enrich a company from its domain. - Verify an email before any separately authorized outreach. This is research infrastructure. It does not replace judgment, qualification, or the revenue queue. It should not become an excuse to build a giant lead database. 4. Competitor and offer research An agent could: - Pull competitor ad examples. - Compare positioning language. - Inspect SEO claims. - Track what vendors publish. - Compare pricing pages and market signals. - Analyze customer comments where the source permits it. That would be useful for full/REFIT offer design and content positioning. 5. Media production experiments The image and video catalog could support bounded experiments in: - Thumbnail variants - Short-form visual tests - Simple explainer assets - UGC-style demonstrations - Creative testing for content packages I would keep this secondary. The current business bottleneck is unlikely to be a shortage of image and video APIs. 6. Sharing selected full/REFIT tools The private registry could eventually hold: - A research skill - A content audit skill - A source-verification tool - A client-safe reporting tool - A small set of approved APIs - A CLI used during delivery That could make the same operating method available across Claude Code, Codex, Cursor, and other agents without copying credentials into each environment. --- What treg does not solve treg does not replace: - A revenue-control queue - Airtable or another system of record - Content approval - Source verification - Editorial judgment - Client permissions - Lead qualification - Human review of sensitive enrichment data - A workflow scheduler - Agent memory - Reliable end-to-end business process design It is an access layer. It makes external capabilities easier to reach. It does not decide whether a call should be made or whether the result is good enough to act on. There is also a risk of making agent behavior more expensive. A model with access to thousands of endpoints can spend money quickly, especially when it retries searches, calls multiple providers, or runs enrichment waterfalls. --- Security and governance considerations The strongest security claim is server-side credential injection. That is useful, but it changes the trust boundary. You would be trusting treg with: - Provider API keys - OAuth tokens - Agent activity - Call metadata - Potentially sensitive enrichment queries - Team membership and usage records Before production use, I would verify: - Secret storage and encryption details - Data retention - Provider terms for scraped or enriched data - Whether query bodies are logged - Whether uploaded skills are scanned - Agent-token scope - Budget enforcement - Revocation behavior - Audit-log completeness - Telemetry settings - Data residency requirements The GitHub repository is public and describes an Apache 2.0 license with additional terms. The additional restriction concerns redistributing the software as a competing hosted registry. Using the hosted treg API inside another product is described as permitted when using its pass-through billing mechanism. --- My recommendation for Paul and full/REFIT [7:01 AM]Recommended: run a small, measured treg pilot rather than installing it everywhere.​ Pilot scope: 1. Create one account and one team. 2. Use the free signup credit. 3. Connect no sensitive production credentials initially. 4. Test three workflows:    - SEO or AEO research    - Public social and YouTube research    - One public-signal company or contact research task 5. Record:    - Actual cost    - Response time    - Provider selected    - Output quality    - Failure rate    - Whether the agent understood the tool correctly 6. Stop if the outputs are generic, unverifiable, or expensive. 7. Expand only if the results reduce real work inside the content or research pipeline. The best first test is not media generation. It is a narrow research job where the alternative would require buying or wiring several provider accounts. The saved research evidence is at: /Users/paul/Documents/Last30Days/treg-agent-tool-registry-raw-v3.md --- Verification Confirmed from the live service and public repository: - Public landing page and /app behavior - Sign-in methods - Catalog and API documentation - llms.txt and skill.md - CLI installation instructions - MCP endpoint - Public use cases and workflows - Server-side credential injection claims - Organization, token, budget, retry, and audit concepts - GitHub repository activity - Recent public sentiment signals Unresolved: - Exact current catalog size - Real-world provider success rates for the specific workflows full/REFIT would use - Data retention and operational security details - Whether treg is materially better than using a smaller set of direct APIs - Actual cost under multi-step agent behavior - Whether the hosted service is reliable enough for production client delivery Sources [1] treg — turn your coding agent into an SEO expert, a media ... https://treg.to/ [2] Treg: OpenRouter for tools with 2,600 APIs, 0% markup ... https://www.producthunt.com/products/treg-openrouter-for-tools [3] API reference — call any tool through one endpoint | treg https://treg.to/docs [4] Resources: what your agent can do with treg.to https://treg.to/resources [5] GitHub - superdesigndev/treg: OpenRouter for agent tools. Join ... https://github.com/superdesigndev/treg [6] Tool catalog — 3,274 API endpoints your agent can call | treg https://treg.to/catalog [7] treg.to — Remote MCP Server https://mcpmarket.com/server/tregto [8] Treg https://selfhosttools.com/tools/treg/ [9] Treg Review 2026: OpenRouter for Agent Tools - toolworthy.ai https://www.toolworthy.ai/tool/treg [10] TREG on the App Store https://apps.apple.com/us/app/treg/id1438251491 [11] Treg - Tutorial https://treg.to/tutorial