Resources
Changelog
Every version of Forgent with its new features, improvements, and fixes. Updated with every significant deployment.
July 2, 2026 — On-chain pivot: Solana wallets, $FORGENT points, Fable 5
- Sign-In With Solana: accounts are gone. You connect a wallet (Phantom, Solflare, Backpack…), sign a free message, and that's your identity. No email, no password — Google login is retired.
- Creator points: publishing to the marketplace (+50) and completed public usage (+2 per run) now accrue points in an on-ledger history, convertible to $FORGENT at token launch. Balance and activity live in Settings → Creator rewards. See the token doc.
- Favorites: a new favorite button on marketplace agent pages, with self-favorites rejected and farm-proof accounting (unique per wallet, reversible, anonymous signals excluded).
- Claude Fable 5 support: Anthropic's latest and most capable model is now selectable in the Builder and filterable on the marketplace — for the hardest, longest-running agents. See the models doc.
- Free Builder and deploy: designing, deploying, and running your agent are all funded by the platform. No API key, nothing to plug in.
May 28, 2026 — Public runnable URL + "Built with Forgent" badge
- Anonymous public trials: a new toggle in the share dialog that lets a visitor without an account run your agent in 1 click from the URL
forgent.me/agent/<id>. The response arrives as a live stream (word by word). See the dedicated doc. - Example prompt: an optional field you fill in on the /deployside that appears pre-filled in the visitor's textarea. Conversion lever #1 — without it, people leave.
- 3 quota safeguards: 3 trials / IP / agent / day + 50 / agent / day + a platform-wide global cap ~$5/day (configurable via
FORGENT_PUBLIC_DAILY_CAP_USD). 429 / 503 rejections with clear messages to visitors. The global cap is sized at ~30% of the Anthropic budget so it never starves members' sessions. - "Built with Forgent" badge: a footer on all shared agent pages, with a link to the landing tagged
?ref=public-agentfor funnel analytics. - Dedicated OG card: when you share the link on Twitter / LinkedIn / Slack / Discord, the preview card shows "Come try [agent name]" in large type, with your creator handle and a "Free 1-click trial" badge. More clickable than the generic OG.
- Future-proof architecture: the
lib/public-billing.tslayer isolates the resolution of "who pays for this run". Today the platform funds every run; the abstraction lets that policy evolve by changing one line, the rest of the code is agnostic. - Anonymized audit log: a new
PublicRunEventtable logs each trial (successful or rejected) with sha256(IP) — no plaintext IP stored. Enables abuse detection without breaking privacy. - OG wordmark fix: the broken "Forg ent" on the root share card is fixed (Satori broke the kerning at the color change, fixed with 2 sibling spans + negative marginLeft).
May 27, 2026 — Infra hardening (migrations, Sentry, rate limit, lineage, tests)
- Versioned Prisma migrations: all schema changes since late April (marketplace, fork, notifications) are consolidated into an idempotent baseline migration. The Vercel build now runs
prisma migrate deployautomatically. No more blind db push, complete history in Git. - Sentry integrated: 3 configs (client/server/edge) + an
instrumentation.tshook + alogError()helper to wrap critical catches. WithoutSENTRY_DSNin the env, it's a graceful no-op. Can be enabled in 5 minutes for production. - Upstash rate limiting on 5 critical endpoints: fork (10/60s), profile PATCH (10/60s), install+view (60/60s per IP), notifications (300/60s read, 10/60s write). Blocks bots that would try to inflate the counters or brute-force username uniqueness.
- Indestructible lineage: each fork now captures a snapshot of the parent's info (slug + name + owner handle). If the original agent is deleted later, the fork keeps a Forked from X · archived chip instead of silently losing the lineage.
- Playwright + GitHub Actions CI: 8 smoke tests on the public marketplace + profile routes. CI runs typecheck + lint on every PR. A safety net before every major refactor.
May 26, 2026 — Complete marketplace overhaul (16 phases)
- New architecture across 4 surfaces:
/marketplacebecomes a 2-choice landing (Templates / Skills),/marketplace/templatesand/marketplace/skillsare the dedicated feeds with a filter sidebar + URL-stateful search bar, and/marketplace/[slug]renders an agent's detail page with hero, system prompt, variables, and stats. - Skills (formerly Boosters) renamed: we're going back to the term "Skill" (more intuitive) for the attachable expertise modules. Backward compatibility maintained on the API side.
- Complete publish flow: a Publish to marketplace button on
/deploy/[id]. 3-state modal: claim username if missing → type/category/slug form → result with copyable URL + external link + unpublish button. - Real DB fork with a counter incremented in a transaction + notification to the original creator. Friction-light fork modal: pre-filled name + optional what are you changing? field.
- Public creator profile:
/u/[username]with a hero (colored avatar + bio + member since), 4 stats blocks (published agents, cumulative installs, forks received, total views), grid of published agents. Anti-enumeration in place (neutral 404 for external visitors on a username with no publications). - Notifications inbox: a bell in the desktop and mobile header with an unread badge. Discreet 60s poll + refresh on focus. Hook in the fork API to notify the owner when their agent is forked.
- My Agents reworked into 3 tabs: Created (originals, no parent), Installed (forks, with a butter forkedchip), Workflows. Public stats (installs/forks/views) visible on a published agent's page.
- Dynamic OG images + dynamic sitemap: each
/marketplace/[slug]and/u/[username]has its own 1200x630 image generated by Satori. The async sitemap reads the DB to add all published agents and creators to the XML. - Install/view/fork counters with client-side sessionStorage dedup + server-side IP rate limit to avoid artificial inflation.
May 20, 2026 — Cross-device persistence + Skills + cost estimates
- Auto-saved drafts in DB: your work in the Builder is saved server-side every ~1.2s (debounced). Open the Builder on your phone, your draft is there. Local sessionStorage stays active in parallel for zero latency.
- Blueprint PDF export: a new "Download PDF" button next to Copy JSON / cURL. Opens a print-friendly window with the D2 logo and your structured sections, then triggers the browser's print dialog. Zero added dependencies — OS-native rendering.
- Persisted version history: the blueprint snapshots (created at deployment) are now in Postgres instead of localStorage. Cap of 20 snapshots per agent per user. The Version history section on the agent page now lists them cross-device.
- Saved workflows: you can save a chain of agents with a name and reload it in 1 click. The Workflows tab lists your saved workflows above the builder. Persisted in DB.
- Skills marketplace V1.5: 6 curated Skills (Financial analyst, Code reviewer, Competitive monitoring, Email writer, Meeting summary, GDPR auditor) at /marketplace/skills. A primary "Start a new agent" button + a secondary "Add to my current draft" button that merges the Skill into your in-progress draft (system prompt appended, tools/MCP deduplicated, variables merged).
- "Edit this agent": a new button on a deployed agent's page that loads the blueprint into the Builder with the "(edited)" suffix. Deploying creates a new version (the original agent stays active until manually archived).
- Cost estimation: the model selector in the Advanced Builder shows "Claude Fable 5" with a per-run estimate. The agent page shows the same estimate. Calculated for an average run of 2k input tokens + 1.5k output. Real-time tracking comes later.
- D2 "Playful no-code" redesign: the entire site moved to a light theme with the extended palette (terra, sage, sky, butter, rose), 1.5px ink borders, black offset sticker shadows on the cards, pill buttons, butter highlights behind emphatic words. Header centered across 3 columns, two robots wandering around the hero and fleeing your cursor.
May 19, 2026 — Navigation overhaul
- New Marketplace hub: the old "Templates" tab has become a /marketplace section that will soon also host attachable Skills and community publications. The 5 existing templates are now at
/marketplace/templates. Details → - Workflows moved to a My Agents tab: the standalone
/workflowspage is gone — it's now a tab inside My Agents, where it conceptually belongs (chaining agents you already own). - Simplified nav: down from 5 items to 3 (+ CTA).Builder · My Agents · Marketplace · Documentation — much more readable for a beginner.
- Old-link compatibility: the old URLs
/templatesand/workflowsremain accessible via HTTP 308 redirect — your bookmarks and shared links keep working.
May 19, 2026 — Advanced Mode, Templates & Workflows
- Simple / Advanced toggle: a new switch at the top of the Builder's right panel. Simple mode keeps the existing behavior (read-only panel, fed by the chat). Advanced mode unlocks everything below. The choice is persistent and synced across pages. Details →
- Direct blueprint editing: in Advanced mode, all fields (name, description, model, tone, system prompt) become editable inputs. Details →
{{x}}variables: placeholders in the system prompt, substituted at export and deployment. Automatic detection of undefined variables with one-click add. Details →- "Test the agent" dry-run: a section to test your prompt without deploying. A direct call to
/v1/messages, response in 2-3s. Details → - A/B testing: compare two prompt variants side by side on the same question, with an Adopt variant B button. Details →
- Version history: each save of a blueprint pushes a snapshot into
localStorage(up to 20). A Restore button that opens the version in the Builder. Details → - Detailed logs: in Advanced mode, an additional panel on the deployment page with the raw stream of SSE events (timestamp + type + full JSON). Details →
- Webhooks & personal API tokens: generate an
fgt_…token in Settings and trigger your agents from the outside viaPOST /api/agents/[id]/run. Details → - Marketplace templates: a new /marketplace/templates page with 5 agents ready to remix (Token Launch Strategist, Wallet Alpha Analyst, Pump.fun Page Builder, Airdrop Points Designer). Details →
- Multi-agent workflows: a new /dashboard?tab=workflows page that orchestrates from the browser the sequential execution of N agents — one's text output becomes the next one's input. Details →
May 18, 2026 — Rebrand & public launch
- New name & domain: EasyAgents becomes Forgent and moves to forgent.me. The old
easyagents-murex.vercel.applink remains accessible but is no longer the official version. - Logo & visual identity: a new wordmark "Forgent" with the orange asterisk, an SVG favicon, and a dedicated OpenGraph image for clean previews when you share the link on LinkedIn, Twitter, iMessage, etc.
- Step-by-step onboarding: new accounts see a checklist on My Agents (design an agent → deploy it) that walks them through their first agent without leaving the site.
May 7, 2026 — Documentation & polish
- Documentation section: everything you're reading right now. 14 pages covering the concepts, the product, and resources, with a sticky sidebar and page-to-page navigation.
- "My Agents": the old "Dashboard" tab has a more meaningful name. The page itself hasn't changed.
- Landing bumped up a notch: a hero that fills the screen, an inline Claude logo next to the title, and a new "Why an agent?" section to concretely explain what it's for.
- Desktop header: full-width navigation instead of being crammed in the middle.
May 6, 2026 — Mobile & measurement
- Mobile responsive: the Builder now has a tabbed pattern on mobile (Conversation / Preview), the Header collapses into a hamburger menu, and the whole site is comfortably usable from a phone.
- Vercel Analytics & Speed Insights: to measure traffic and performance in production. Anonymous data on the visitor side.
April 24, 2026 — First launch
- Forgent (then EasyAgents) launches in beta on
easyagents-murex.vercel.app. - Google Auth + encrypted server-side storage (AES-256-GCM) of sensitive credentials.
- Builder: a 5-step guided chat that builds an agent plan in parallel on the right.
- Direct deployment to Anthropic Managed Agents via the
/api/ma-proxyproxy. - My Agents: a list of your deployed agents with status, JSON import/export of plans.
For what's coming next, see the Roadmap.