Changelog

Versioning & Release Policy

TameFlare follows Semantic Versioning (semver):

  • Major (X.0.0) — breaking changes to API, CLI, or configuration format. At least 30 days deprecation notice before removal.
  • Minor (0.X.0) — new features, new doc pages, new connectors. Backward-compatible.
  • Patch (0.0.X) — bug fixes, security patches, documentation updates. Backward-compatible.

Release cadence

| Type | Frequency | Examples | |---|---|---| | Patch releases | As needed (typically weekly) | Bug fixes, security patches | | Minor releases | Monthly | New features, new connectors | | Major releases | When necessary (rare) | Breaking API changes |

Deprecation policy

  • Deprecated features are marked in the changelog and docs with a Deprecated badge
  • Deprecated features continue to work for at least 2 minor releases (typically 2 months)
  • Removal is announced in the changelog with migration instructions
  • Breaking changes are always documented in the Upgrade Guide

Staying updated

v2.0.2 — Security & Trust Hardening

Based on a 150+ question new-user audit across 4 personas (solo engineer, security engineer, engineering manager, SRE).

Security

  • SECURITY.md — vulnerability disclosure process, security design principles, known limitations, supported versions
  • CA key protected — ca.key and ca.crt added to .gitignore (auto-generated per installation, must never be committed)
  • CLI token security — tf connector add now supports --token-stdin (pipe from stdin) and --token-env <VAR> (read from env var) to avoid shell history exposure. --token flag warns about history visibility

Landing Page

  • CTA clarity — "Open dashboard" → "Self-host now" with "Requires local install" subtext
  • Zero telemetry statement — added to "What TameFlare is not" section: no analytics, no phone-home, no license checks
  • Connector badges — now labeled "(all stable)" with action counts for all 5 connectors
  • Version badge — "v2.0 · Production-ready" in hero area

Configuration

  • .env.example — POSTHOG_KEY and SENTRY_DSN commented out with "disabled by default, zero telemetry" note

Purchase Flow & Community (v2.0.11)

  • Support (new page) — 4-tier support matrix (Community/Pro/Team/Enterprise), response time SLAs by severity (critical/high/medium/low), email support scope table, enterprise support details (dedicated Slack, named engineer, professional services), self-help resource links, security vulnerability reporting process
  • Plans page expanded — purchase flow FAQ (manual process with 24h SLA, self-serve planned), payment methods table (credit card, invoice NET 30, wire transfer), 30-day money-back refund policy, annual pricing (coming soon ~20% discount), overage behavior (hard block at limit, 402 response, monthly reset at UTC), enterprise contact details
  • Changelog expanded — versioning & release policy (semver commitment, release cadence table, deprecation policy with 2-release notice period, staying updated via GitHub releases)
  • Security page expanded — third-party audit statement (honest: not yet audited, planned), test coverage table (278+ tests across 6 suites), security-specific test areas (auth bypass, RBAC, kill switch, rate limiting, nonce replay, Zod validation)
  • About page expanded — team & governance section (maintainers, governance model, bus factor with honest assessment), security audit status summary
  • Sidebar expanded — Support page in Reference, Community section (GitHub Discussions + Source Code with external link handling), Learn section (tutorials, example agent repo)
  • CONTRIBUTING.md — contribution guidelines, code style, testing instructions, license agreement

Team Rollout & Exit Documentation (v2.0.10)

  • Team Rollout (new page) — multi-environment deployment guide (separate instance per env, shared policies via export/import, single-instance alternative), RBAC reference (4 roles, 12-row permissions matrix, honest "no custom roles"), user registration flow (first user = owner, REGISTRATION_OPEN, DEFAULT_USER_ROLE), new member onboarding, multi-tenancy status (honest: single-org per instance), SSO roadmap (planned for Team tier), agent-scoped visibility (current state + roadmap), config migration guide (8-step checklist)
  • Uninstall page expanded — vendor lock-in assessment (6-component risk table, removal path per integration mode, transferable concepts)
  • Plans page expanded — per-environment licensing FAQ, downgrade procedure (resource-by-resource impact table with what happens to excess agents, actions, audit, and features)

Audit, Monitoring & Operations Documentation (v2.0.9)

  • Audit Log (new page) — append-only architecture (no delete API, honest note about no cryptographic chaining), 26 event types with full schema, event-specific JSON examples, per-tier retention (30d/30d/1yr), cleanup job (manual + cron), CSV + webhook export, SIEM integration guide (Splunk, Datadog, ELK, Sumo Logic), compliance evidence guide (approval chain proof, 5 provable claims, 3 honest limitations)
  • Architecture page expanded — component independence (gateway runs without control plane), gateway health monitoring (traffic log discrepancy detection, silent drop detection), log rotation & disk management (growth estimates, manual prune, thresholds), HA/scaling limitations (honest: single-node SQLite, no HA gateway, Turso for multi-node control plane, planned improvements)
  • Policies page expanded — policy hit analytics (SQL query for top policies, pattern diagnosis table, dashboard charts, Prometheus PromQL)

Integration & Connector Documentation (v2.0.8)

3 new documentation pages + integration guide expanded:

  • Connectors — per-connector reference: GitHub (20+ actions, GHE custom base URL), OpenAI (15+ actions, Azure workaround via generic connector), Stripe (11+ actions), Slack (10+ actions), Generic HTTP (method-based, action counting, custom auth header). Custom connector development guide with Go interface, Jira example, and registration instructions.
  • Proxy Behavior — headers (added/stripped/preserved/not-logged), SSE streaming (fully supported), chunked transfer, protocol support matrix (HTTP/HTTPS yes, WebSocket/gRPC/UDP no with rationale), NO_PROXY domain exclusions, upstream proxy chaining (not yet supported + workaround), connection handling (timeouts, pooling, keep-alive), rate limiting reference (120 req/min proxy, 60 req/min SDK).
  • Observability — 10 Prometheus metrics with labels, scrape config, 6 Grafana PromQL dashboard panels, 6 Prometheus alert rules (high denial rate, kill switch active, gateway down, high latency, agent rate limited, upstream errors), Alertmanager integration (Slack + PagerDuty), Datadog options, health monitoring endpoints.
  • Integration guide expanded — framework interception explanation (transport-layer, 9 framework compatibility table), multi-agent setup guide (DevOps Bot + Finance Bot example, agent isolation, naming conventions).

Policy & Enforcement Documentation (v2.0.7)

Policies page expanded from 572 to 910 lines with 7 new sections:

  • Policy model overview — access rules vs policies, which takes precedence, evaluation order (kill switch → access rules → policies), when to use each (decision table)
  • Decision algorithm — deny-wins precedence with pseudocode, 5-row precedence table, worked example with 3 policies
  • Default behavior — no matching policy = deny with hint, catch-all policy example with warning
  • Policy scoping guide — 6 scope levels (global, per-connector, per-action-type, per-agent, per-environment, combined), priority range recommendations
  • Policy-as-code workflow — git → PR → CI → API deployment, GitHub Actions example, version history in dashboard
  • Risk model — 6 built-in risk hints, agent-provided vs connector-provided, risk scoring table (0-100)
  • Testing expanded — dry-run response example (JSON), dashboard UI walkthrough, simulation endpoint for batch testing

Core Reference Documentation (v2.0.6)

9 new documentation pages added to the "Operations" section:

  • Failure Modes — gateway crash, SQLite corruption, approval timeout (5min → 408), DNS failure, malicious YAML handling, disk full, connector auth errors, cold start times, load balancer compatibility
  • Backup & Restore — SQLite backup (file copy + .backup API), config export/import, full backup script, automated cron, disaster recovery matrix (5 scenarios)
  • Upgrade Guide — standard upgrade (git pull + pnpm db:push), version pinning, breaking change checklist, v1→v2 migration steps, Docker Compose upgrades, rollback procedure
  • Performance — proxy latency (p50/p95/p99), policy evaluation scaling, gateway resource usage, control plane resource usage, SQLite throughput, cold start times, scaling limits
  • Architecture — component diagram, 3 deployment topologies (single-node, split, Docker Compose), network requirements, blast radius analysis, scaling patterns
  • CI/CD Integration — GitHub Actions workflow, Docker-based CI, ephemeral container pattern, health check best practices, artifact collection
  • Database & Storage — why SQLite, full schema overview (17 tables), data growth estimates, Turso cloud support, data residency matrix
  • Uninstall & Exit — 6-step removal guide, credential extraction, data export, system trust cleanup, GDPR data deletion
  • Kill Switch Recovery — activation (dashboard/CLI/API), investigation runbook (7 steps), notification options, false alarm prevention

Docs sidebar reorganized with new "Operations" section.

Landing Page & Pricing Copy (v2.0.5)

Landing Page

  • Agent definition — hero now explains "any process that makes HTTP calls — Python scripts, LangChain agents, n8n workflows, shell scripts"
  • "Zero code changes" clarified — "(Mode A)" qualifier added; CLI section explains Mode B requires SDK
  • Platform template badges — OpenClaw, LangChain, n8n, Claude Code all marked "(stable)"
  • Monitoring scope — "What TameFlare is not" clarifies: monitors own activity only, not app metrics/infra/APM
  • ES256 link — "Allow" decision links to Concepts docs page for token details
  • Fail-closed link — "Deny" decision links to Security docs page
  • Security model summary — new section: HTTPS interception, credential isolation, fail-closed, link to full security docs
  • Social proof — GitHub stars button, "source-available since 2025", TameFlare vs alternatives link
  • Roadmap link — added to footer navigation
  • Demo sandbox — CTA section links to dashboard demo sandbox

Pricing Page

  • Scenario table — "Which plan fits your team?" with Solo dev / Small team / Platform team recommendations
  • 4 new FAQs — agent limit (total vs concurrent), license expiry (falls back to Starter), annual pricing (not yet), trial (Starter = full eval)
  • Action counting — clarified: only connector-matched requests count, blocked-no-connector don't
  • Purchase SLA — CTA buttons now say "Get license key (within 24h)"
  • Trial statement — Starter description: "All features testable — no time limit"

Security Model Documentation (v2.0.4)

  • HTTPS interception explained — TLS termination with local CA, per-installation uniqueness, trust store configuration (Python, Node.js, Go, macOS)
  • Proxy bypass analysis — honest limitations table (5 bypass methods) with OS-level mitigations
  • Credential vault deep dive — encryption details, access matrix (5 actor types), key loss recovery
  • Data logging scope — what tf logs (11 data types) and what it intentionally does NOT log (5 types with reasons)
  • Inter-component security — localhost-only default, service token auth, agent port isolation
  • Data residency — zero outbound calls table, Slack notification data scope, no telemetry confirmation
  • Fail-closed model — 9 denial scenarios documented (no fail-open mode)
  • Operational checklist expanded — added signing keys, CA key, credential vault permissions
  • Responsible disclosure — updated email, linked to SECURITY.md

First-Run Experience (v2.0.3)

  • Prerequisites section — Node.js ≥18, pnpm ≥8, Go ≥1.22, OS support matrix (Linux/macOS/Windows native)
  • pnpm requirement — explicit explanation that npm/yarn won't work with monorepo workspaces
  • pnpm dev scope — clarified: starts Next.js control plane only, gateway started separately
  • Smoke test — new step 5: curl /api/health expected output + tf status
  • Docker Compose services — lists both services started (web + gateway) with ports
  • Config file guide — .TameFlare/ directory contents with commit safety table (config.yaml = safe, ca.key = never)
  • Environment variables — split into Required (none for dev), Recommended (4 for production), Optional
  • ES256 key auto-generation — explained: ephemeral for dev, persistent for production with link to env-vars page
  • Port conflict troubleshooting — PORT=3001 pnpm dev and .TameFlare/config.yaml port override
  • System requirements — Gateway ~20-50 MB RAM, control plane ~200-400 MB RAM
  • CLI token security — --token-env GITHUB_TOKEN in quickstart examples
  • Env-vars docs — telemetry section clarified as "disabled by default, zero telemetry"

v2.0 — Proxy-First Architecture

Major release: transparent HTTP proxy that intercepts all agent outbound traffic. Zero code changes required.

Gateway v2 (Go)

  • Transparent HTTP/HTTPS proxy — intercepts all agent outbound traffic via HTTP_PROXY/HTTPS_PROXY
  • Per-agent proxy ports — each agent gets a dedicated port for traffic isolation
  • ECDSA P-256 CA — auto-generated TLS certificate authority for HTTPS interception
  • SQLite traffic logger — every request logged with agent, domain, action, decision, latency
  • Enforcement levels — monitor (log only), soft_enforce (log would-deny), full_enforce (block)

Connectors (6 built-in)

  • GitHub — 20+ action types (PRs, issues, branches, releases, files, repos, webhooks, workflows)
  • OpenAI — 24+ action types (chat, completions, embeddings, images, audio, files, fine-tuning, assistants, batches)
  • Anthropic — messages, models (dual-provider with OpenAI connector via provider field)
  • Stripe — 40+ action types (charges, refunds, payment intents, customers, subscriptions, invoices, payouts, transfers, products, prices, disputes)
  • Slack — 35+ action types (messages, channels, files, reactions, pins, users, bookmarks, reminders, admin ops, webhooks)
  • Generic HTTP — any domain, bearer/api_key/basic auth

Permission System

  • Per-agent, per-connector, per-action permissions with wildcard patterns (github.pr.*)
  • 5-level resolution — exact match > wildcard > all actions > global > default deny
  • Deny-all default — no connector configured = no access

Credential Vault

  • AES-256-GCM encrypted credential storage on disk
  • Proxy injects credentials at request time — agent never sees real API keys

Approval Workflow (Proxy)

  • Hold-connection model — proxy holds HTTP connection open until human approves (5 min timeout)
  • Approve/deny via CLI (tf approvals approve/deny) or dashboard
  • SQLite-backed with waiter channels for goroutine synchronization

Scoped Kill Switch

  • Block ALL traffic, or scope to a specific connector or agent
  • Instant activation via CLI or dashboard

CLI v2

  • tf init — scaffold .TameFlare/ directory, generate TLS CA, start gateway
  • tf run --name "Bot" <cmd> — register process, set proxy env vars, spawn process
  • tf status, tf stop, tf logs — gateway management
  • tf connector add/list/remove — manage connectors
  • tf permissions set/list — manage per-agent permissions
  • tf approvals list/approve/deny — manage pending approvals
  • tf kill-switch activate/deactivate — emergency controls

Platform Templates

  • tf init --platform openclaw — OpenAI + Anthropic, deny all else
  • tf init --platform langchain — LLM APIs + search tools
  • tf init --platform n8n — OpenAI + Slack + GitHub workflows
  • tf init --platform claude-code — GitHub + package registries

Dashboard Updates

  • Traffic page — live proxy traffic log with decision badges, method colors, agent/decision filters
  • Connectors page — live connectors from gateway, available types, permission model docs
  • Proxy approvals — one-click approve/deny with 5s auto-refresh
  • Overview — proxy gateway status, agent/connector counts, traffic stats

Landing Page & Docs

  • Landing page rewritten for proxy-first model with terminal-first code examples
  • README rewritten with proxy quickstart, architecture diagram, connector docs
  • Public docs updated: quickstart, concepts, API reference, integration guide, changelog

Hardening (v2.0.1)

  • Rate limiting — 120 req/min per agent, sliding window, 429 with Retry-After
  • Prometheus metrics — GET /internal/metrics with aaf_traffic_total, latency histograms, active agent/connector gauges, uptime
  • Response headers — X-TameFlare-Decision, X-TameFlare-Agent, X-TameFlare-Connector, X-TameFlare-Action on all proxied responses
  • Structured JSON logging — optional JSON log output to stderr with agent, connector, decision, latency fields
  • SQLite WAL mode — journal_mode=WAL + busy_timeout=5000 for concurrent performance
  • Docker — apps/gateway-v2/Dockerfile (multi-stage Alpine), docker-compose.yml updated (gateway-v2 primary, v1 legacy profile)
  • CI — gateway-v2 build + vet, CLI typecheck, v1 gateway renamed to legacy

v0.8.1 — Devil's Advocate Improvement Pass

Comprehensive improvement pass based on a 200-question devil's advocate audit of all user-facing pages, dashboard, and documentation.

Trust & Credibility

  • Version badge (v0.8.0) added to landing page hero and About page
  • ELv2 license tooltip in footer with plain-English explanation
  • About page rewritten with project maturity metrics, component status, and license section
  • "Coming soon" labels replaced with "In progress" + trackable issue links

Landing Page Clarity

  • Code snippet extended to show full request → execute flow (8 lines)
  • Latency note added: "Policy evaluation typically completes in single-digit milliseconds"
  • Mode B description improved: "Ideal for initial rollout and monitoring"
  • Connector claims qualified: "GitHub (11 actions) and generic webhooks"
  • "No managed/hosted version is planned" added to "What TameFlare is not"
  • Problem cards softened: "few or no runtime guardrails" instead of "no guardrails"

Pricing & Purchase Flow

  • Slack approvals and webhook connector moved to Starter tier (free)
  • Starter audit retention increased from 7 → 30 days
  • Policy Builder and email approvals removed from paid tiers (not implemented)
  • New FAQ: "Is the Starter tier really free?"
  • Upgrade link added to Settings > Plan section

Dashboard Onboarding & UX

  • Settings page reordered: Plan & License first, kill switch second
  • Enforcement level recommendation text added (L0 → L1 → L2 progression)
  • Production checklist improved: severity indicators (Critical/Recommended/Optional) + generation commands
  • Integration links: Slack setup, GitHub PAT creation, Docker Compose gateway URL hint
  • Agents empty state: suspend vs revoke explanation, runtime field description, lost key guidance
  • Policies: "recommended for new users" on GitHub Safe Defaults pack, improved priority explanation
  • Approvals: sandbox hint in empty state, approver group tooltip
  • Connectors: Gateway start command when offline
  • Webhooks: webhook_url usage explanation in empty state
  • Users: Copy invite link button, DEFAULT_USER_ROLE env var documentation
  • DEFAULT_USER_ROLE env var implemented (default: viewer, configurable)

Documentation

  • New Environment Variables reference page with all env vars, defaults, and generation commands
  • Quickstart env var table expanded with 5 new variables
  • Plans docs updated: Starter includes Slack + webhook + 30-day audit
  • Concepts: default role text updated to mention DEFAULT_USER_ROLE
  • Troubleshooting: Turso env var names fixed (TURSO_DATABASE_URL)
  • Sidebar: Environment Variables page added to Reference section

v0.8.0 — Messaging, Security & UX Improvements

Comprehensive improvement pass based on a 206-question audit of all user-facing pages.

Messaging & Trust

  • Standardized licensing language to "source-available under ELv2" across all pages
  • Removed all cloud/managed service references (no cloud product exists)
  • SDK install commands updated to install-from-source (npm/PyPI packages coming soon)
  • Honest connector count: 2 built-in (GitHub + Webhook), not implied broader support
  • Qualified data privacy claims with integration caveats

Registration & Security

  • Registration closed by default after the first user (bootstrap mode auto-opens)
  • New REGISTRATION_OPEN env var to allow new signups when needed
  • Encryption at rest ungated to all tiers — security is no longer a paid feature
  • Register page shows closed state with admin instructions

Landing Page

  • New Two integration modes section: Mode A (full enforcement) vs Mode B (policy-only)
  • New "What TameFlare is not" callout: not prompt engineering, not monitoring, not a framework, not cloud

Dashboard UX

  • Getting Started button — re-open the welcome guide from the overview page
  • Production readiness checklist on Settings page (6 live checks)
  • Improved agents page empty state with API key guidance
  • Policy model explanation (deny-wins, priority, scopes) on policies page
  • Sandbox description clarified: no real execution, policies auto-installed on first run

Documentation

  • Fixed wrong license in FAQ (was "MIT", now ELv2)
  • Fixed priority explanation that contradicted deny-wins model
  • Fixed stale "Managed Cloud" changelog entry
  • Updated plans docs: encryption listed under Starter tier

Legal

  • Terms of Service updated: ELv2 license, removed managed service sections
  • Privacy Policy updated: "source-available" language, self-hosted data handling

v0.7.0 — Licensing & Paid Tiers

Tiered licensing model

  • 4 tiers: Starter (free), Pro ($20/mo), Team ($49/mo), Enterprise (custom)
  • License keys are HS256-signed JWTs with tf_lic_ prefix
  • Set TF_LICENSE_KEY env var or activate via dashboard/CLI
  • No internet required for license validation (offline HMAC verification)

Usage tracking & enforcement

  • Monthly action counters per org, tracked in usage_counters table
  • Agent limits enforced at creation time
  • 402 Payment Required when plan limits reached
  • 403 Forbidden for feature-gated endpoints

Feature gates

  • Batch API (Team+), Slack approvals (Pro+), webhook connector (Pro+)
  • Prometheus metrics (Pro+), config export/import (Team+)
  • Encryption at rest (all tiers as of v0.8.0), webhook callbacks (Pro+)
  • Audit retention enforced per tier

Dashboard plan UI

  • PlanUsage component with progress bars and tier badge
  • PlanBanner upgrade CTA at 80%+ usage
  • FeatureGate locked overlay for gated features
  • New Settings > Plan page with feature comparison table
  • Sidebar footer shows current tier badge

License management

  • POST /api/license/activate — validate and activate a key
  • GET /api/license/activate — check current license status
  • POST /api/license/issue — generate signed keys (admin, MAINTENANCE_SECRET)
  • /activate page — UI for pasting license keys
  • CLI: tf license status, tf license activate, tf license usage

SDK updates

  • Node.js + Python: 402/403 errors are not retried (immediate throw)
  • Error objects include upgradeUrl, tier, feature, requiredTier
  • Python: is_plan_limit and is_feature_gated convenience properties

Pricing page rewrite

  • Starter / Pro / Team cards with Enterprise CTA
  • Updated FAQs for license key model
  • Source-available language replaces open-source

Documentation

  • New Plans & Licensing page in public docs
  • TECHNICAL_SPEC Section 13: Licensing & Tiers
  • OpenAPI spec updated with license endpoints
  • README updated with ELv2 license, new env vars

v0.6.0 — Improvement Tracker Complete

Batch action endpoint

  • New POST /api/v1/actions/batch — submit up to 20 action requests in a single call
  • Each action evaluated independently against the same policy set
  • SDK: new batchRequestActions() method in Node.js SDK
  • Documented in API Reference with request/response examples

Prometheus metrics endpoint

  • New GET /api/metrics — Prometheus text format for observability integration
  • Metrics: total actions, active agents, hourly rate, 24h outcome breakdown, audit event count
  • Protected by MAINTENANCE_SECRET bearer token
  • Compatible with Grafana, Datadog, and any Prometheus scraper

SDK improvements

  • Failover mode (circuit breaker): failoverMode: 'open' returns a synthetic deny result when TameFlare is unreachable instead of throwing
  • onFailover callback for alerting/logging when failover triggers
  • Batch requests: batchRequestActions() for multi-action evaluation
  • Publish prep: npm exports map, keywords, repository, engines, README for Node.js; PyPI classifiers and project URLs for Python

Parameter redaction

  • Sensitive fields in action parameters (password, token, secret, api_key, authorization, etc.) are automatically redacted with [REDACTED] before database storage
  • Original values used for policy evaluation and token hashing but never persisted
  • New redactSensitiveFields() utility in @/lib/redact

Framework integration guides

  • New section in SDK docs: LangChain (Python), OpenAI function calling (Node.js), CrewAI (Python)
  • Shows the universal pattern for wrapping tool calls with TameFlare policy checks

Migration guides

  • Added to Troubleshooting docs: SQLite → Turso, password → email auth, adding encryption at rest, upgrading between versions

Documentation expansion

  • New Action Types page (/docs/action-types) — all 16 Gateway actions, naming convention, custom types
  • New Security page (/docs/security) — threat model, token lifecycle, encryption, auth, operational checklist
  • Performance & architecture section in Deployment docs — architecture overview, latency, scaling, resource requirements
  • Cloud deployment guides — Fly.io, Railway, AWS ECS with step-by-step instructions
  • 10 new FAQ entries — hallucination, deny-wins, token timing, failover, GDPR, timezones
  • About page (/about) — project status, motivation, community links

Landing page & pricing

  • Hero section jargon clarified with plain-English explanations
  • GitHub stars badge and community links added
  • Misleading claims fixed (latency, connector scope, setup time)
  • Non-Slack approval options documented on landing page
  • Pricing page rewritten: vaporware tiers removed, honest Starter / Pro / Team / Enterprise model

Dashboard UX

  • 6 demo scenarios across 3 domains (GitHub, payments, infrastructure)
  • Policy Builder wizard — 5-step form alternative to raw YAML editing
  • Auto-refresh toggle — visible Live/Paused button with localStorage persistence
  • Agent management — suspend action with tooltip explanations, better UX
  • RBAC visibility — role descriptions and invite hint on Users page
  • Connectors page — Gateway optional note, expandable action lists per connector
  • Configurable per-agent rate limits — checkRateLimit() accepts optional maxRequests

v0.5.0 — Credibility, Integration & Onboarding

Integration guide

  • New Integration Guide docs page with end-to-end walkthrough for connecting agents to TameFlare
  • Two integration modes documented: Mode A (full enforcement via Gateway) and Mode B (policy-only)
  • Framework-specific examples: Node.js, Python, LangChain, CrewAI, and raw HTTP
  • Step-by-step credential migration path from agent to Gateway
  • Recommended rollout plan (L0 Monitor → L1 Soft enforce → L2 Full enforce)
  • Troubleshooting table for common issues

Custom sandbox

  • Sandbox now has Presets and Custom tabs
  • Custom tab provides a JSON editor with a starter template
  • Type any action spec and evaluate it against your installed policies
  • Results show outcome, risk score, reason, and matched policies

Welcome modal

  • First-time users see a 4-step guided tour on their first dashboard visit
  • Steps: try sandbox, install policies, register agent, connect agent
  • Dismissible via "Get started" button, remembered in localStorage

User management

  • New Users page at /dashboard/users with role badges and status indicators
  • Role change dropdown (owner/admin/member/viewer) with RBAC enforcement
  • Suspend and activate controls for user accounts
  • POST /api/dashboard/users endpoint protected by admin role
  • Three new audit event types: user.role_changed, user.suspended, user.activated

Contextual empty states

  • Actions page empty state now shows icon, explanation, and links to sandbox and integration guide
  • Approvals page empty state explains requires_approval and links to policies and Slack config
  • Risk score indicator shows Low/Medium/High label and risk factor explanation on hover

Credibility fixes

  • Created /privacy and /terms pages (previously 404)
  • Fixed placeholder GitHub repo URLs (your-org → agentfirewall) across README, quickstart, and docs
  • Changed pricing CTA from "Start free trial" to "Get started"
  • Aligned enforcement level names across Settings UI, Concepts docs, and Technical Spec (Monitor / Soft enforce / Full enforce)
  • Removed unused env vars (NEXTAUTH_SECRET, GITHUB_CLIENT_ID/SECRET), documented MAINTENANCE_SECRET and AUDIT_RETENTION_DAYS
  • Fixed footer links on landing and pricing pages
  • Fixed stale "NextAuth" references in Technical Spec (actual auth is custom email/password with bcrypt + DB sessions)

Inline help links

  • Settings integrations form: links to Slack app creation, GitHub token creation, and Gateway setup guide
  • Create Policy modal: "Syntax docs" link to Writing Policies documentation

Production deployment guide

  • New Deployment docs page covering Docker Compose, environment variables, key generation, SQLite vs Turso, TLS/HTTPS (Nginx, Caddy, cloud), platform guides (Fly.io, Railway, VPS), approval channels, maintenance, and backups

Encryption key warning

  • Settings integrations form shows an amber warning banner when SETTINGS_ENCRYPTION_KEY is not configured
  • Explains that secrets are stored in plaintext without the key

Toast notifications for settings

  • Enforcement level, organization settings, and integration settings all show success/error toast notifications on save
  • Uses the existing useToast() system

Password reset

  • Admins can reset user passwords from the Users page via the Manage dropdown
  • Generates a temporary password shown once in a modal with a copy button
  • Login page "Forgot password?" link explains the self-hosted reset process

Troubleshooting & FAQ

  • New Troubleshooting docs page with 9 common issues (auth errors, policy matching, gateway connection, encryption, login, DB upgrades) and 9 FAQ entries

Your First Policy tutorial

  • New step-by-step tutorial: define a policy goal, write YAML, test in the sandbox, understand rule evaluation, and refine
  • Covers deny, requires_approval, and allow rules with real examples

Connector action documentation

  • API Reference now lists all 11 GitHub connector actions and 5 webhook connector actions with parameter tables

Server-side audit search

  • Audit log search now runs server-side via SQL LIKE queries across event type, agent ID, action type, and details
  • Previously filtered client-side after fetching all results — now pagination works correctly with search

Action type filter

  • Actions page now has an action type search input alongside the existing status filter buttons
  • Active filters shown as removable chips with a "Clear all" option

Trust signals

  • GitHub repo icon link added to landing page navigation bar
  • Footer updated with GitHub link and MIT License badge

v0.4.0 — Dashboard UX & Observability

Agent detail page

  • New Agent Detail Page at /dashboard/agents/[id] — click any agent name to see the full profile
  • 7-day activity sparkline (Recharts bar chart), action history table with risk scores, stats (total actions, today, denied, pending approval)
  • Agent info card with runtime, environment, API key prefix, status, timestamps
  • Audit timeline showing every event for that agent
  • Inline controls: suspend, activate, revoke — with toast notification feedback

Policy version history

  • New policy_versions database table tracks every change to a policy (version number, YAML content, who changed it, change type)
  • New Policy Detail Page at /dashboard/policies/[id] — click any policy name to see current YAML, tags, priority, and full version timeline
  • Versions are recorded automatically on policy creation (update tracking coming next)

Command palette

  • Press Cmd+K (Mac) or Ctrl+K (Windows/Linux) to open a global search overlay
  • Search across 8 dashboard pages and 4 quick commands (toggle kill switch, register agent, create policy, export audit)
  • Full keyboard navigation: arrow keys to move, Enter to select, Escape to close
  • Search button with platform-aware shortcut hint in the dashboard header

Toast notifications

  • New ToastProvider context and useToast() hook for mutation feedback
  • Three types: success (green), error (red), warning (amber)
  • Auto-dismiss after 4 seconds with slide-out animation
  • Wired into agent detail controls (suspend/activate/revoke)

Breadcrumb navigation

  • Auto-generated from the current URL pathname
  • Clickable segments for every level (Dashboard > Agents > agent-id)
  • Label map for all known dashboard pages, truncation for long IDs
  • Hidden on the overview page to keep it clean

Responsive mobile layout

  • Sidebar hidden on screens below lg breakpoint
  • Hamburger menu button in the header opens a slide-out panel with full sidebar navigation
  • Responsive padding (px-4 on mobile, px-8 on desktop)
  • Overlay backdrop when mobile menu is open, auto-closes on navigation

Webhook delivery log

  • New webhook_deliveries database table logs every webhook callback attempt
  • fireWebhook() now records URL, HTTP status code, latency (ms), response body, and errors
  • New Webhooks page at /dashboard/webhooks with stats cards (total, successful, failed) and a delivery history table
  • Added to sidebar navigation under Setup

Dark/light theme toggle

  • New ThemeProvider context with useTheme() hook supporting dark, light, and system modes
  • Persisted to localStorage, respects prefers-color-scheme media query
  • Toggle appears in the UserMenu dropdown with Moon/Sun/Monitor icons
  • System mode automatically follows OS preference changes

Config export/import

  • Export: GET /api/dashboard/export (requires admin) — downloads all policies, agents, and org settings as a JSON file
  • Import: POST /api/dashboard/import (requires owner) — restores policies and agents from a JSON bundle, creates audit trail entry
  • Imported agents receive new API keys; policies are created as new entries

API response headers

  • All /api/* responses now include X-Request-Id (unique per request) and X-TameFlare-Version headers
  • Injected via Next.js middleware on both public and dashboard API routes
  • Middleware matcher expanded from dashboard-only to all API routes

v0.3.0 — Platform Maturity

Action investigation

  • New Action Detail Page at /dashboard/actions/[id] — click any action row to see the full lifecycle
  • Action spec JSON, policy evaluation trace (matched policies, risk factors, constraints), decision details, approval flow, execution result
  • Audit timeline showing every event for that action in chronological order

Analytics dashboard

  • Recharts-powered charts on the Overview page: 7-day stacked area chart (actions over time by outcome), pie chart (decision breakdown), horizontal bar charts (top agents by volume, top denied action types)
  • Charts appear automatically once you have action data

Shared UI components

  • StatusBadge, RiskIndicator, and StatCard extracted to @/components/ui/ and deduplicated across dashboard pages
  • Consistent styling and behavior everywhere these components appear

Real health monitoring

  • New GET /api/health endpoint that checks database connectivity and gateway latency
  • HealthStatus client component in the dashboard header replaces the previously hardcoded "All systems operational"
  • Hover tooltip shows per-service status with latency in milliseconds
  • Auto-refreshes every 30 seconds

Input validation

  • Zod schemas for all major API routes: action requests, dry-run, agent creation, approval responses
  • POST /api/v1/actions now returns structured validation errors with field paths
  • Error format: { "error": "Validation failed", "details": "action_spec.resource.provider: provider is required" }

RBAC enforcement

  • New requireRole() middleware with role hierarchy: owner > admin > member > viewer
  • Kill switch toggle requires owner role
  • Agent and policy management requires admin role
  • Returns 403 Forbidden with required_role and your_role in the response

Data cleanup

  • New POST /api/maintenance/cleanup endpoint for purging expired data
  • Cleans expired token nonces, expired sessions, and used nonces older than 24 hours
  • Optional audit event retention via AUDIT_RETENTION_DAYS environment variable
  • Protected by MAINTENANCE_SECRET bearer token

Query layer hardening

  • All 13 database query catch blocks now log errors via logQueryError() instead of silently swallowing them
  • Organization ID cache uses a 60-second TTL instead of caching permanently (fixes stale cache in multi-org scenarios)

Policy editor

  • Create Policy modal now has a line number gutter alongside the YAML editor
  • Live YAML validation via the dry-run API (debounced at 500ms) with inline feedback
  • Valid/invalid indicator next to the editor label
  • Error and warning panels below the editor with specific messages
  • Full TameFlare policy template as placeholder text

Hardcoded values fixed

  • Sidebar plan label reads from org.plan in the database instead of hardcoded "Free plan"
  • Landing page footer links now point to /privacy, /terms, and mailto: instead of #
  • Health status in dashboard header is real, not decorative

v0.2.0 — Auth, Testing, and Developer Experience

User authentication

  • Email/password registration and login with bcrypt-hashed passwords
  • Session tokens with secure cookie management
  • User menu in dashboard header with role display and logout
  • Backward-compatible with DASHBOARD_PASSWORD shared password mode
  • First registered user automatically becomes org owner

Integration settings

  • New Settings > Integrations section for Slack, GitHub, and Gateway configuration
  • Credentials saved to DB and read at runtime via getIntegrationSettings()
  • Env vars take precedence; DB settings used as fallback
  • "Test connection" buttons for Slack and Gateway
  • Settings cache with 10s TTL, invalidated on save

Demo sandbox

  • Three built-in scenarios: safe (create issue), risky (merge PR), destructive (delete branch)
  • "Try the Firewall" buttons on dashboard overview
  • Real policy evaluation with audit trail

Audit log improvements

  • Search input filters events by text across all fields
  • Event type dropdown filters by specific event types
  • CSV export with optional type and search filters

Dashboard polish

  • Interactive enforcement level selector (L0/L1/L2) persisted to DB
  • Dynamic org name in sidebar from DB instead of hardcoded value
  • Connectors page "Configure" buttons link to Settings > Integrations

Testing

  • 51 policy engine tests via vitest
    • 35 condition tests covering all 14 operators, nested combinators, empty groups
    • 16 evaluator tests covering deny-wins, scope filtering, priority, risk scoring, traces

Infrastructure

  • GitHub Actions CI pipeline: typecheck, unit tests, build (web + docs + gateway)
  • All hardcoded localhost URLs replaced with env-driven config (NEXT_PUBLIC_DOCS_URL, NEXT_PUBLIC_APP_URL)
  • Updated .env.example with all new variables

v0.1.0 — MVP

Initial release with full E2E flow.

Core

  • Policy engine with YAML DSL, deny-wins precedence
  • ES256 decision tokens with nonce replay protection
  • 13 API routes including dry-run and token verify
  • In-memory rate limiting (60 req/min per agent)
  • Webhook callbacks on action decisions
  • Structured JSON logging with request IDs
  • Multi-org support

Gateway

  • Go execution gateway with 11 GitHub action handlers
  • Generic webhook connector (5 HTTP methods)
  • Real token verification via control plane

Dashboard

  • 8 pages with real-time auto-refresh
  • Policy dry-run UI, interactive kill switch, approve/deny buttons
  • Paginated tables, trend indicators, error boundaries
  • Live gateway connector status
  • Password protection (opt-in)

SDKs and CLI

  • Node.js SDK with auto-retry and approval polling
  • Python SDK (sync + async)
  • CLI: init, status, agents, policies commands

Infrastructure

  • Docker Compose (web + gateway + seed)
  • 3 built-in policy packs
  • OpenAPI spec