Forge takes a rough use-case description, iterates it through your agent framework, and delivers a battle-tested skill file — no prompt engineering degree required.
Tell Forge what you want in plain language. A multi-turn interviewer extracts the real requirements.
Baseline, variate, score, iterate. Each cycle generates N variations refined from your feedback.
Download a production-ready skill file, drop it into your agent, and run.
Everything you need to ship reliable agent skills without the guesswork.
Conversational interview that extracts structured specs from vague ideas.
Configurable breadth (width) and depth — try 3 variations, iterate 3 cycles.
Score variations 0–10, pick a winner, and Forge iterates toward it.
Winning skills are saved to a library. Reuse, remix, share across projects.
Create, monitor, and retrieve jobs programmatically. MCP server for agentic control.
Redis-backed priority queue. Run 2 parallel jobs by default, scale up as needed.
One product, three tiers. Pick what fits your volume.
For individuals exploring
For power users & teams
For heavy automation
All plans include the same core optimization loop. Tiers limit volume and parallelism only.
Every feature is available via REST API. Use the MCP server for agentic orchestration.
curl -X POST https://forge.myforge.work/jobs \
-H "Content-Type: application/json" \
-d '{
"use_case": "Write meeting notes from a transcript",
"target_framework": "claude-code",
"width": 3,
"depth": 3
}'