One file in. Judgment on demand.
Forecry sits between your agent and its riskiest moments. The agent asks, a hosted specialist answers, and the answer comes with reasons and a receipt.
Four steps, no rewrite.
Subscribe and get a key
Pick a plan, check out through Stripe, and create an API key from the dashboard. Keys are shown once.
Install one connector file
Forecry generates the connector for your runtime — Claude Code, Codex, Hermes, or generic HTTP/MCP. Drop it in; nothing else changes.
$ forecry install release-guardian --target claude-code
Your agent asks before acting
When the agent hits a decision a behavior covers — a deploy, a dependency add, a migration — it consults the hosted specialist and gets GO or NO-GO with reasons.
Read the receipts
Every consultation is logged: trigger, evidence, rule applied, verdict, credits spent. The dashboard shows what your agent was told and why.
The parts people ask about.
01What runs locally vs. hosted?
The connector file is the only local piece. Decision policies, evidence handling, and verdict logic run on Forecry's side, so updates land without you reinstalling anything.
02Can the behavior take actions itself?
No. Behaviors return verdicts and plans. Your agent (and its own permission model) still does the acting — Forecry adds the judgment, not another pair of hands.
03What does a consultation cost?
Each consultation spends credits from your monthly plan. Light checks cost less than deep ones, and every receipt shows its exact credit cost. Overage is metered per credit at your plan's rate.
04What happens if Forecry is down?
Behaviors are hosted, so if Forecry is unreachable your agent simply doesn't get a verdict and falls back to its own defaults. The status page is public and stays reachable even when you can't sign in.