First public open-source release of the OpenAgentic platform
First public open-source release of the OpenAgentic platform
Release history, source-derived from version.json — the single source of truth for platform versions.
“Open Field” 2026-06-04 (current)
First public open-source release of the OpenAgentic platform
One-command install (install.sh) — Docker Compose stack with the Ink setup wizard
Chat with streaming, smart model routing, and zero hardcoded model IDs
9 built-in MCP servers (aws, azure, gcp, kubernetes, prometheus, loki, github, admin, web)
Flows workflow engine with ops Flow templates and human-in-the-loop gates
Immutable audit trail + server-side approval gate on mutating tool calls
Per-user cross-mode memory shared by chat and flows
Admin console: providers, MCP fleet, audit log, dashboards
Zero telemetry — no phone-home, ever
“Light It Up” 2026-04-10
Admin Test Harness (Bolt 7) — SSE-streaming system test suite for K8s, MCP, agents, chat
Artifact / diagram agent upgrade — production-quality HTML+CSS+JS, Mermaid removed
Agent synthesis enforcement — raw JSON blocked from reaching users; Ollama format extraction
Azure MCP creation tools — VNet, subnet, NSG, Application Gateway, Front Door verified end-to-end
Smart router sentinel fixes across the chat pipeline
Model Registry rebuilt from live SDK discovery; staleness banner removed
TTFT optimization — Ollama warm-up + parallel MCP/Memory initialization
Bob UC1–UC5 all pass against live Azure; Bolt 4 full 24/24 workflow regression with 0 failures
Security: hardcoded JWT secret removed from test harness; token-cost warning in UI
“Lockstep” 2026-04-08
Lockstep platform tagging — agentic, openagentic, oat, openagentic-sdk all at v0.6.2
Dual-mode build pipeline — vendored or git-clone-via-PAT companion sources
AIF Claude as a first-class provider with Anthropic-native SSE streaming
Live-SDK Model Registry — Admin LLM Registry rebuilt from provider SDK discovery on every load
Smart router sentinel fix in the chat validation stage
Self-correction pipeline + ReAct loop with final-synthesis fallback
Cloud_operations agent type with Azure OBO token propagation
Prisma singleton sweep — fewer pg connections under load
Memory prompt-injection sanitization (SEV0) + provider-disable-cascades-to-models (SEV0)
Openagentic-manager image no longer pulls from GAR at runtime — bundled at build time
Model Registry no longer persisted as a curated list; discovered from each provider SDK on load
“Overhaul” 2026-03-10
Flows Enterprise Overhaul — Per-node/workflow/admin config hierarchy with 3-level governance
Unified Cross-Mode Memory — Chat and flows share a single per-user context layer
Error Handling Framework — Error edges, handler nodes, and error routing in workflow engine
Workflow Versioning — Auto-snapshot on save, visual diff, version restore
Agent Platform — Persona system, tool policy, visual cron scheduling
Admin Dashboard Overhaul — Recharts charts + AdminMetricCard/AdminFilterBar across all 10 admin views
LLM Provider Setup Wizard — 4-step first-time configuration wizard for fresh deployments
Slack & Teams Integration — Inbound webhook → workflow execution → Block Kit/Adaptive Card responses
SetupWizard component for zero-config first-time deployment
DeploymentStateService with cached state detection and setup completion
9 new admin API endpoints (deployment-state, pause/resume, model disable/enable, capability-matrix, credential rotation)
SlackIntegrationService with signature verification and Block Kit formatting
TeamsIntegrationService with Bot Framework token verification and Adaptive Cards
IntegrationMessageFormatter for workflow output to platform-specific rich messages
Integration CRUD routes with webhook receivers and connection testing
IntegrationsView admin UI with cards, logs, add/edit dialog
LLMProviderManagement rewrite with provider cards, detail panels, capability matrix view
Recharts AreaChart/BarChart/PieChart added to PerformanceMetrics, EmbeddingMetrics, MonitoringView, ContextWindowMetrics, UserActivityDashboard, OATUsageStatsView, PromptMetrics, DashboardOverview
AdminMetricCard with sparklines and trends across all admin views
AdminFilterBar replacing manual select/filter UI in all views
Legacy auth migration (useAuth/getAuthHeaders → apiRequest) in PerformanceMetrics, MonitoringView
Prisma schema: LLMProvider status/paused_until/disabled_models/capability flags, Integration + IntegrationLog models
Per-node advanced config: disabled, timeout, retry, error routing, pinned output, notes
Per-workflow settings: model defaults, cost budgets, env vars, tags, visibility
Admin workflow governance: 5-tab panel with execution limits, cost caps, model restrictions, node type controls, memory settings
Enhanced agent node config: persona, tool policy, budgets, approval, memory scope
Error handler node type with log/retry/notify/transform actions
User context node type for cross-mode memory access in workflows
Error edge type with red dashed styling and error output handles
Workflow versioning: auto-snapshot on save, version history sidebar, visual diff, restore
UserContextService: unified cross-mode memory assembly from chat, workflows, memories
7 new API endpoints for user context CRUD, admin stats, workflow governance
Auto-indexing: chat messages and workflow results indexed into unified context
Per-user memory settings: source toggles, retention, context-in-chat/code toggles
Admin memory dashboard: user context stats, search, purge, retention config
Agent persona system: role, tone, boundaries, bootstrap instructions
Agent tool policy: allow/deny modes with high-risk auto-detection
Agent scheduling: visual cron builder, target workflow, input templates
3-level config hierarchy: admin caps > workflow defaults > node overrides
Fixed #22c55e (apple puke green) replaced with #00D26A terminal green across 25+ files
Fixed --color-success CSS variable from #30D158 to #00D26A
Fixed legacy auth patterns in PerformanceMetrics and MonitoringView using deprecated useAuth hook
Fixed broken memory.stage.ts — prisma.memory model doesn't exist, replaced with UserMemoryService
Fixed memory stage async fire-and-forget — now synchronous (awaited) before completion stage
Fixed SSE event streaming — nodes show RUNNING state in real-time (socket.setNoDelay + flush)
Fixed MCP error detection — properly unwrap proxy response layers (proxy → JSONRPC → MCP result)
Fixed openagentic-proxy CrashLoopBackOff — duplicate /metrics route registration
Fixed chat.openagentic.local 404 — TLS proxy routing to wrong backend
Fixed Prisma schema — missing Integration, IntegrationLog models and LLMProvider extended fields
Added UserMemoryService — unified memory read (Redis→Milvus→PG→merge), write (PG+Milvus+Redis)
Added UserProfileService — user personality learning with style_dna via LLM
Added FeedbackLearningService — feedback→preferences→system prompt injection
Added LLM-powered memory compaction (topic clustering, summarization, profile recompute)
Added per-user model preferences in SmartModelRouter
Added /api/user-memory/* API routes (entries, context, profile, ingest, purge)
Added auto-ingestion hooks (chat, tool, workflow, feedback)
Added ExecutionResultsPanel resize handle (280-800px) and close button
Added MCP tool node retry with graceful degradation
Added pre-upgrade backup and post-upgrade verification scripts
“Hardened” 2026-02-22
DLP Scanner — 50+ detection rules for credentials, PII, infrastructure secrets
Immutable Audit Trail — Cryptographic hash chaining, tamper-evident logs
Mandatory HITL Enforcement — Server-side tool approval gate for high-risk operations
Per-Tool Credential Scoping — Tools only receive credentials they need
Structured Error Classification — Typed failover reasons for LLM providers
Pipeline Hook System — 16 extensible hook points for security and observability
Agent Proxy — Multi-agent orchestration with parallel, sequential, and supervisor patterns
Apple-inspired color palette — Real greens, blues, purples across the UI
DLPScannerService with scan/redact/block actions
ToolApprovalGate for server-side human-in-the-loop enforcement
CredentialScopeService for per-tool credential isolation
FailoverError typed error classification for LLM providers
Shared retry framework with configurable backoff policies
Request deduplication cache
SSE event sequencing with gap detection
Service lifecycle management with ordered startup/shutdown
Pipeline hooks system with modifying, void, and sync hooks
Agent Proxy service for multi-agent orchestration
Session loading bug fix — all message fields preserved on reload
Version bump across all services to 0.5.0
Fixed session loading dropping toolCalls, toolResults, thinkingSteps
Fixed Anthropic content block arrays rendering as raw JSON
Fixed neon colors (#00FF00, #FF0000) replaced with Apple palette
Fixed version endpoint showing stale version info
“Titan” 2026-02-13
Platform Hardening & Security - 75 CVE patches, auth bypass removal
Unique colorful Admin Console icons with gradient animations
Tool result grounding workflow with LangGraph integration
Improved streaming code block handling
K8s service discovery audit and port fixes
11 new unique colorful SVG icons for Admin Console sidebar
Incomplete code fence handling during streaming
Tool result grounding service with async processing
Comprehensive K8s service discovery validation
Fixed Admin Console sidebar showing folder icons for all sections
Fixed code blocks breaking live streaming rendering
Fixed LangGraph URL port mismatch (8002 -> 3100)
“Cartographer” 2026-01-18
Live artifact streaming - artifacts render as they stream, not after completion
Flowise Google auth fix - workspace lookup via WorkspaceUser entity
Unified version system across all components
StreamingArtifactRenderer component for live artifact preview
Detailed JWT logging in Flowise middleware
Fixed TypeORM EntityPropertyNotFoundError for Flowise workspace lookup
Fixed Admin Console Flowise opening user account instead of system@internal
“Pioneer” 2026-01-10
Smart model routing for automatic model selection
MCP Tools integration
Real-time workspace sync with MinIO/S3
Multi-provider LLM support (Anthropic, OpenAI, Google, Azure, Ollama)
“Genesis” 2025-12-01
Initial release
Chat mode with streaming
Basic MCP proxy
SSE streaming chat
Azure AD and Google OAuth
Flowise workflow integration