Share an agent
Forgent lets you share any deployed agent via a public URL. Visitors can see the agent's profile and, if you allow it, run it in 1 click without even creating an account. It's the viral engine of the product.
The two sharing modes
When you publish an agent from /deploy, you get a unique URL of the form forgent.me/agent/<shareId>. This URL can work in two different modes.
Mode 1 — Read-only sharing
Enabled by default when you click Make public. The visitor lands on a profile that shows:
- The agent's name and description
- The model used (Claude Fable 5)
- The capabilities (web search, files, code, etc.)
- The connected apps (MCP)
- A Remixbutton that copies the agent into the visitor's builder (account required)
This mode is cosmetic: no visitor can run the agent. Ideal for a portfolio, frozen demos, and screenshots.
Mode 2 — Public trials enabled
Toggle Enable public trials in the share dialog. There, the public page gains a launch area with:
- A pre-filled text field with the example prompt you wrote
- A Run the agent button that calls Claude in streaming
- The response that appears live word by word
- A button to try again with another prompt
This is the viral mode. You share the link on Twitter, people click, and they try it in 5 seconds without creating anything.
The example prompt
The Example promptfield in the share dialog is what the visitor sees pre-filled in the textarea. It's your inline onboarding: it implicitly says “here's how to use me”.
A few good examples:
- Pump.fun Page Builder → “Create launch metadata and a pinned tweet for $FROG, a Solana culture coin about on-chain builders”
- Wallet Alpha Analyst → “Summarize this wallet activity and tell me what behavior patterns stand out”
- On-chain Risk Monitor → “Review this token launch data and list the main red flags to check before sharing it publicly”
Without an example prompt, the visitor lands on an empty textarea and might leave without knowing what to type. It's conversion lever #1.
The 3 quota guardrails
To keep a viral agent from draining the platform budget in one night, Forgent applies three cascading limits:
- 3 trials per IP per agent per day.A single visitor can't spam your agent in a loop. On the 4th trial, the message “Create a Forgent account to continue”.
- 50 trials per agent per day, across all visitors. If your agent is a hit, the trials cut off at 50 and resume the next day. The visitor sees a clear message.
- ~$5 per day of public spend across the whole Forgent platform. A global cap that protects your Anthropic budget. Aligned with ~30% of the daily Anthropic budget, the rest is reserved for the sessions of registered members (including you).
The three quotas reset at midnight UTC. The refused trials are still logged in the audit ledger (without storing a plaintext IP — sha256 everywhere) so we can detect any potential abuse.
Security: what's exposed?
The public page never reveals:
- The platform credentials— used only on the server side, never sent to the visitor's browser.
- Your agent's system prompt— it's used to drive Claude on the server side, and the visitor only sees the agent's response.
- The variables you configured as default values.
- The internal IDs of your blueprint (the shareId is a random 96-bit token, non-enumerable).
What the visitor sees: name, description, model, capabilities, the names of the connected apps (just their labels, not the URLs or credentials).
forgent.me. It's what closes the viral loop: the visitor tries your agent, sees the badge, clicks, creates their own account, and builds their own agents.Unpublishing
The Unpublish button in the dialog puts the agent back to private: the shareId is set to null, and the public URL immediately returns 404. If you re-publish later, a new shareId is generated — the old link stays dead for good. This is intentional: it guarantees that someone who saved your URL will no longer be able to use it once it's unpublished.
Tracking public trials
The share dialog shows a counter X public trials total. For more detailed stats (who tried, when, at what cost), a dedicated analytics view will come in a future iteration. In the meantime, thePublicRunEvent table in the DB contains the whole history (anonymized via sha256).
Typical use cases
- Portfolio — you show a client an agent you built for them, without them installing anything.
- Viral growth— you post on Twitter “I built a fact-checker, test it here ” with the link.
- Interactive documentation — you write a blog post about a use case, and embed the playable agent as a demo.
- Team onboarding — you want your colleagues to test an agent before explaining how to create their account. You just share the URL.