Changelog

What's new in FORG

Every release, every feature, every fix. Subscribe to the GitHub releases feed for notifications.

June 3, 2026

Token savings fixed — compaction tracking + expanded model support

  • FixedToken Savings no longer shows 0 — cache savings and context compaction savings are now both tracked and displayed
  • NewContext compaction savings tracked — each PreCompact event records how many tokens were in the context before FORG compacted it
  • NewDashboard 'Tokens Saved' column shows combined cache + compaction savings with a tooltip breakdown
  • NewSession detail Cache Savings card updated to show combined total with per-type breakdown
  • ImprovedModel pricing expanded: minimax-m3, minimax-m1, o3, o3-mini, o4-mini, claude-opus-4-6/4-7/4-8, claude-sonnet-4-6, claude-haiku-4-5 now all have accurate pricing
  • Fixedforg update correctly resolves to v3.1.1 and replaces binary on all 5 platforms

June 3, 2026

Model tracking, adapter parity, and FORG-OS signal layers

  • NewPer-model token breakdown in session detail — see exactly which model used how many tokens, cache reads, and cost per session
  • NewCodex CLI promoted to co-primary adapter — full official support alongside Claude Code with dedicated install path and signal coverage
  • NewGemini CLI adapter now official — MCP server registration in ~/.gemini/settings.json with session telemetry
  • NewFORG-OS signal layer specification (L1 shell, L2 process, L3 OS) — foundation for cross-tool token aggregation
  • FixedTokens Saved now shows correct USD savings (cache_tokens_read × 90% of input rate) instead of raw token count
  • Fixedtokens_in no longer double-counts cache reads — clean split: tokens_in = new input only, cache separately tracked
  • FixedSession model_usage preserved on Supabase outage — KV keys deleted only after confirmed PATCH success
  • FixedIdempotent turn telemetry — duplicate Stop hook fires no longer double-count tokens via ON CONFLICT(turn_uuid) DO NOTHING
  • ImprovedModel names normalized on storage (date suffix stripped) — GROUP BY model correctly merges claude-sonnet-4-5-20250929 variants
  • Fixedforg update no longer returns HTTP 404 — v3.1.0 binaries now confirmed in remote R2 bucket for all 5 platforms
  • ImprovedSessions list shows +N badge when a session used multiple models

May 31, 2025

v3.0.3

All commands verified, dashboard connected on activate

  • Fixedforg activate now starts a session immediately — dashboard shows 'connected' without running forg on
  • Fixedforg done uses active persisted session; no --session flag required
  • Fixedforg set-goal uses active persisted session; error if no session active
  • Fixedforg refocus unpauses session before emitting refocus-ack signal
  • Fixedforg status correctly shows session ID, pause state, and goal
  • Fixedforg deactivate emits session-end before wiping local data
  • New"cli" added as valid ToolName in engine for signals emitted directly by CLI commands
  • ImprovedAll 5 release binaries (darwin-arm64, darwin-amd64, linux-amd64, linux-arm64, windows-amd64) published to R2

May 29, 2025

v3.0.2

Signal routing and session key fixes

  • FixedFORG_INTERNAL_KEY now set on both api.forg.pro and forg-rule-engine workers
  • FixedEngine-to-license-worker S2S session key lookup correctly authorized
  • Fixedforg update correctly replaces binary in-place on all platforms
  • Improvedforg on extracted RunOn() shared function; used by both forg on and forg activate

May 28, 2025

Bug fixes and stability

  • Fixednpm publish pipeline now correctly bundles native binaries for all platforms
  • FixedSession timeout rule no longer fires on sessions with <5 calls
  • FixedDashboard cost totals now correctly exclude refunded signals
  • ImprovedAgent startup time reduced from 340ms to 180ms on macOS
  • FixedWebhook retry backoff now uses exponential strategy (was linear)

May 20, 2025

FORG v3 GA — The AI Control Plane

  • NewComplete rebuild: Go agent, Rules Engine v2, Next.js 15 dashboard
  • NewThree-pillar architecture: Observe, Control, Optimize
  • NewRules engine with budget, model policy, rate limit, and session policy rules
  • NewFORG Atlas alpha: natural language queries over your usage data
  • NewCryptographic audit chain: tamper-evident signal history

Apr 15, 2025

FORG Atlas dashboard and AI insights

  • NewFORG Atlas alpha: natural language usage queries (Professional+ plans)
  • NewAI Insights dashboard tab with automated waste detection
  • NewModel recommendation engine beta
  • ImprovedSignal ingestion throughput increased to 10,000/sec per Worker
  • FixedCache hit rate calculation now excludes cache_write tokens

Mar 28, 2025

Business plan, SCIM, and data residency

  • NewBusiness plan launched: SCIM provisioning, data residency, priority support
  • NewSCIM 2.0 support for Okta, Azure AD, and Google Workspace
  • NewEU data residency option (Business+): signals stored in EU-region only
  • NewDPA (Data Processing Agreement) now available for all Business+ customers
  • ImprovedSAML SSO now supports JIT (just-in-time) user provisioning

Mar 10, 2025

Gateway enforcement layer

  • NewGateway enforcement mode: pre-call rule evaluation (sync, ~3ms P50)
  • NewModel redirect support: block → redirect to cheaper model automatically
  • NewFail-open configuration: gateway unreachable = allow call
  • ImprovedBudget accumulator P99 latency reduced from 45ms to 11ms
  • FixedRolling 30-day window correctly handles month boundaries

Feb 20, 2025

v2.5.0

Team analytics and org hierarchy

  • NewTeam-level cost dashboards with org hierarchy support
  • NewTeam dimension auto-tagging via .forg/project.yaml
  • NewManager view: see all team members' usage in one dashboard
  • Improvedk-anonymity threshold now configurable (minimum 3, default 5)
  • FixedPer-team budget rules correctly scope to team dimension

Feb 5, 2025

v2.4.0

Webhooks and event streaming

  • NewWebhook support: rule enforcement events, budget warnings, anomalies
  • NewSlack integration: native notifications for enforcement events
  • NewServer-Sent Events stream: real-time signal feed (dashboard + API)
  • ImprovedWebhook delivery: exponential backoff retry, 72-hour retention

Jan 22, 2025

v2.3.0

Budget enforcement rules

  • NewBudget rules: per-user, per-team, per-org daily/weekly/monthly limits
  • NewRate limit rules: calls per hour/day with burst support
  • NewSession policy rules: idle timeout, max duration
  • NewWarn-only mode: rules log but don't enforce (for tuning)
  • FixedBudget window rollover now handles DST transitions correctly

Jan 8, 2025

v2.2.0

VS Code and JetBrains adapters

  • NewVS Code adapter: GitHub Copilot + Codeium + Continue.dev support
  • NewJetBrains adapter: AI Assistant and Copilot signal collection
  • NewAdapter auto-detection: forg detect installs the right adapter
  • ImprovedAll adapters now support project.yaml dimension auto-tagging

Dec 18, 2024

v2.1.0

SSO (SAML 2.0 + OIDC)

  • NewSAML 2.0 SSO: Okta, Azure AD, Google Workspace, generic IdP
  • NewOIDC support: any compliant identity provider
  • NewRBAC roles: Owner, Admin, Member, Viewer
  • ImprovedSession key derivation now uses HKDF-SHA256 (was HKDF-SHA1)

Dec 1, 2024

v2.0.0

Rules engine GA

  • NewRules engine v1 GA: model policy rules in production
  • NewModel allowlist and denylist rules with glob pattern support
  • NewAudit log v1: every rule evaluation recorded
  • ImprovedSignal schema v2: adds latency fields (ttft, total)

Nov 15, 2024

v1.9.0

Cursor and Windsurf adapters

  • NewCursor adapter GA: full signal collection including Tab completions
  • NewWindsurf adapter beta: Cascade session tracking
  • ImprovedCursor adapter correctly attributes cost to project via workspace detection

Nov 1, 2024

v1.8.0

API keys for Team+ plans

  • NewAPI key management: create, rotate, revoke programmatic API keys
  • NewAPI key scopes: read-only, write, admin
  • NewCI/CD integration: API keys for automated signal ingestion in pipelines

Oct 15, 2024

v1.7.0

k-anonymity enforcement

  • Newk-anonymity ≥ 5 enforced on all individual developer reports
  • NewAutomatic aggregation: teams <5 members get org-level reports only
  • ImprovedPrivacy dashboard: shows what data is visible at your team size

Oct 1, 2024

v1.6.0

Claude Code adapter GA

  • NewClaude Code adapter GA: all 7 hook event types
  • NewSession tracking: session-level cost and token aggregation
  • Newstatusline.sh integration: FORG cost shown in Claude Code status bar
  • FixedHook installation no longer overwrites pre-existing custom hooks

Sep 15, 2024

v1.5.2

Performance and stability

  • FixedSignal queue flush now correctly handles offline reconnection
  • ImprovedWorker memory usage reduced 40% via signal batching
  • FixedDashboard cost totals now refresh correctly after rule changes

Sep 1, 2024

v1.5.0

Supabase pgvector integration

  • Newpgvector extension: signal embeddings for FORG Atlas (future feature)
  • NewSignal search: find sessions by model, user, cost range
  • ImprovedSignal retention configurable: 30d, 90d, 1yr (by plan)

Aug 15, 2024

v1.4.0

Dashboard beta

  • NewWeb dashboard beta: cost over time, model breakdown, session list
  • NewCost by model chart: daily and weekly views
  • NewSession explorer: drill into individual session cost and timing

Aug 1, 2024

v1.3.0

Multi-model support

  • NewSupport for OpenAI GPT-4 family (gpt-4o, gpt-4-turbo)
  • NewSupport for Google Gemini 1.5 family
  • NewModel pricing database: auto-updated cost per token by model

Jul 15, 2024

v1.2.0

Session attribution

  • NewSession ID tracking: group all calls in a session for session-level cost
  • NewUser dimension: attribute cost to individual developers
  • NewProject dimension: attribute cost to git repository / project