Control

Total Control. Zero Compromise.

A complete control plane for AI usage across your org. Budget enforcement, model policy, and team-level governance — cascading from org down to every individual session.

3
Scope levels
< 2ms
Policy eval
Git-native
Rules format
Real-time
Enforcement
Start controlling Read the docs

Governance that cascades

Rules flow from org-level down. Lower scopes can only restrict — never expand — what the level above allows.

Org
Managed by: Security & Compliance team
Global model blocklist · Compliance mandates · Spend ceilings · Geo restrictions
Visibility
All teams & users
Team
Managed by: Team leads / eng managers
Per-team budgets · Model preferences · Time windows · Usage caps
Visibility
Team members only
Individual
Managed by: Developers (within team limits)
Personal spend view · Session history · Cannot exceed team policy
Visibility
Own sessions only

Budget enforcement, three ways

Choose how aggressive your guardrails are. Mix modes across teams.

Hard stop
Session is blocked the moment the budget ceiling is hit. No exceptions, no grace period. Protects against runaway agents and batch jobs that ignore soft limits.
When: 100% of budget consumed
Soft cap
Usage continues past the threshold but is flagged and reported. The user sees a warning banner. Great for tracking without blocking productivity.
When: Configurable % — typically 90%
Alert
Silent notification to a Slack channel, email, or webhook when spend crosses a threshold. No disruption to the user. Used for early-warning dashboards.
When: Any configurable $ or % threshold
Team Budget Status — May 2025
Engineering$3,840 / $5,000
77% used$1,160 remaining
Product$1,120 / $2,000
56% used$880 remaining
Design$890 / $1,000
89% used$110 remaining
Total org spend$5,850 / $8,000

Control which models your teams can use

Allowlists, blocklists, and approval flows — applied at policy eval time, not at the API.

Scenario A: StartupCost focus

Early-stage teams allowlist only GPT-4o-mini and Claude Haiku. Powerful enough for 90% of tasks, at a fraction of flagship cost. Any request for a blocked model is auto-redirected.

gpt-4o-mini✓ allowed
claude-haiku-3-5✓ allowed
gpt-4o✗ blocked
claude-opus-4✗ blocked
Scenario B: EnterpriseCompliance

Large orgs blocklist experimental and preview models. Production traffic uses only approved, stable versions. New models require a formal approval workflow before they're unblocked.

gpt-4o-2025-04✓ approved
claude-sonnet-4✓ approved
gpt-4o-preview⏳ pending
gemini-experimental✗ blocked
policy.forg.yaml — model policy block
# Model policy — Enterprise org
model_policy:
mode: blocklist
blocked_patterns:
- "*-preview"
- "*-experimental"
new_model_approval: required
approvers: ["security-team", "cto"]

Four dimensions of control

Cover every enforcement scenario without writing custom middleware.

Budget Rules
Set hard stops, soft caps, and alert thresholds per user, team, or org. Never get surprised by an invoice again.
Model Rules
Allowlist or blocklist specific models. Require approval workflows before new models reach production.
Time Rules
Restrict AI usage to business hours, enforce cooldowns, or block overnight batch jobs from runaway.
Usage Rules
Cap tokens per session, rate-limit per user, and kill runaway sessions before they drain your budget.

Start with a template

Drop a proven policy in with one command. Customize from there.

StartupCost-first

$500/mo cap · Claude Haiku only · Business hours only

Hard stop at $500/mo org-wide
Claude Haiku as sole allowed model
Active 09:00–18:00 weekdays
Slack alert at 80% spend
TeamBalanced

$5K/mo cap · Approved models · No overnight usage

$5,000/mo team budget
Pre-approved model allowlist
Block sessions 22:00–06:00
Weekly spend digest to manager
EnterpriseFull control

Custom caps · Full allowlist management · 24/7 with alerts

Per-team custom budgets
Centrally managed allowlist
24/7 with anomaly alerts
Immutable audit log
Security-firstCompliance

Model blocklist · Geo restrictions · Mandatory session logging

Block all non-approved models
Geographic access controls
Mandatory session recording
Automated compliance reports

Rules as code. Rules in git.

Export your entire policy set to a YAML file, commit it to version control, and import it anywhere. Rules are diffable, reviewable, and rollbackable — just like application code.

Full audit trail on every policy change
PR reviews for rule updates
Instant rollback with git revert
Sync rules across environments
terminal
$ forg rules export > .forg/rules.yaml
✓ Exported 12 rules to .forg/rules.yaml

$ git add .forg/rules.yaml
$ git commit -m "tighten budget caps"
[main a3f92c1] tighten budget caps

$ forg rules import .forg/rules.yaml
✓ Applied 12 rules — active in 1.8ms

Before FORG. After FORG.

Before — Ad-hoc AI usage
✗Engineers choose any model freely
✗Costs discovered in end-of-month invoices
✗No rollback on policy changes
✗Compliance by honour system
After — Controlled environment
Policy-gated model access, enforced at eval time
Budget alerts fire before overspend occurs
Rules versioned in git, rollback in one command
Automated compliance enforcement with audit log

Take back control of your AI stack

Set your first policy in minutes. Ships with sensible defaults — tighten or loosen any rule from the dashboard or API.

Start controlling View policy docs