Enterprise AI platforms are only as capable as the runtimes they execute. If the runtime doesn't expose audit events, traces, policy decisions, identity, workflow context, and security controls, every platform has to rebuild them. Forge is building those runtime primitives. Forge v0.16.0 extends the runtime with AWS Bedrock support, platform admission hooks, distributed trace propagation, workflow correlation, and enterprise security controls. 🚀 Forge v0.16.0 ☁️ AWS Bedrock Support — native SigV4 outbound authentication for Anthropic and OpenAI-compatible endpoints
🛡️ Platform Admission Hooks — enable AI platforms to enforce quotas, budgets, and execution policies before an agent runs
🔗 End-to-End Trace Propagation — subprocesses and external tools automatically join the same distributed trace
📊 Expanded OpenTelemetry — authentication, scheduling, channel delivery, and admission are instrumented out of the box
🔄 Workflow Correlation — separate workflow definitions from executions for complete lifecycle visibility
🌐 Automatic Workflow Context Propagation — securely propagate workflow identity across trusted internal services
✏️ Skill Builder Edit Mode — iterate on existing skills directly from the web dashboard
🔐 Unified Audit & Redaction Pipeline — one consistent security model across audit logs, guardrails, and traces The goal isn't to build another agent framework. It's to provide the runtime capabilities that enterprise AI platforms need to operate agents securely, observe them end to end, enforce policy, integrate with enterprise infrastructure, and govern them at scale. That's what Forge is building. Full release notes → go.useforge.ai/rel-v0160 #Forge #EnterpriseAI #AIAgents #AgentRuntime #OpenTelemetry #AmazonBedrock #Kubernetes #OpenSource

Most AI agent frameworks stop at inference. Enterprise agents need much more: 🛡️ Security controls that can be audited 📅 Scheduling that survives process restarts 🏢 Multi-tenant governance 🔍 Compliance evidence for every decision ☸️ Native cloud infrastructure integration That's what Forge v0.15.0 delivers. Forge v0.15.0 🚀 ☸️ Kubernetes-Native Scheduling — deploy agent schedules as native Kubernetes CronJobs 🔐 End-to-End Guardrails Audit — every input, context, tool, output, and stream decision is now auditable 📡 OpenTelemetry Guardrail Spans — blocked prompts and policy violations show up directly in traces 🏢 Multi-Tenant Audit Stamping — org_id, workspace_id, entity_id, and entity_type on every audit event 🔍 SIEM-Ready Governance — filter audit streams by organization, workspace, or agent without joins 🤖 forge auth — token management and Kubernetes secret generation built in 📚 Forge as Go Modules — embed forge-core, forge-skills, and forge-plugins directly into your platform One of the biggest additions: Forge agents can now be scheduled using native Kubernetes CronJobs. No custom scheduler. No database dependency. No timing drift. The Kubernetes control plane serves as the scheduler, etcd as the persistence layer, and Forge automatically generates CronJobs, RBAC, and deployment artifacts. Enterprise agents need governance as much as intelligence. Forge now provides out-of-the-box auditability across guardrails, traces, tenants, workspaces, and agents. Full release notes → go.useforge.ai/rel-v0150 #Kubernetes #AIAgents #AgentOps #OpenSource #OpenTelemetry #Security #PlatformEngineering

Enterprise AI agents can’t be governed with logs alone. When an agent makes 10 LLM calls, invokes 5 tools, talks to 3 other agents, and takes action in production, you need to answer: 🔍 Why did it do that? 💰 What did it cost? 🛡️ Was it compliant? 📊 How did it perform? 🔗 Which workflow triggered it? That’s why distributed tracing isn’t observability theater—it’s the foundation for enterprise agent governance. Forge v0.14.0 🚀 📡 End-to-End OpenTelemetry Tracing for AI Agents 🔗 Multi-Agent A2A flows appear as a single distributed trace 🤖 Automatic tracing for LLM calls, tool execution, and outbound APIs 📊 GenAI semantic conventions for token usage, model attribution, and cost analysis 🛡️ Audit ↔ Trace correlation with trace_id and span_id on every audit event 🏛️ Governance-ready visibility for security, compliance, evaluations, and operations ☸️ OTLP collector auto-added to egress policy and Kubernetes deployments ⚡ Works with Tempo, Jaeger, Honeycomb, Datadog, Grafana Cloud, and any OTLP backend The best part? No manual instrumentation. Forge agents are automatically instrumented end-to-end—from the inbound A2A request to every model call, tool invocation, downstream agent hop, and external API request. One trace. Complete lineage. Production ready. 45 files changed. 5.5k+ lines added. 8 issues closed. Full release notes → go.useforge.ai/rel-v0140 #OpenTelemetry #A2A #AIAgents #AgentOps #Observability #OpenSource #useforge #initializ

Forge v0.13.0 is where Forge’s enterprise runtime story becomes verifiable: A2A conformance, hardened audit contracts, platform policy, cancellation, rate limits, and SIEM-ready stream separation. Forge v0.13.0 🚀 🧩 A2A 0.3.0 Agent Card conformance 🔐 Hardened audit pipeline with schema_version + monotonic seq 📡 Audit export via Unix Domain Socket + HTTP fallback 🏛️ Three-layer platform policy: system / user / workspace 🚦 Configurable rate limits built for orchestrated workloads 🧾 Token usage + duration headers for cost enforcement 🛑 tasks/cancel now actually cancels in-flight invocations 🔀 Ops logs to stdout, audit NDJSON to stderr 🤖 New Claude knowledge skills for Forge + Skill Builder 119 files changed. 16k+ lines added. 13 issues closed. Forge’s security and observability claims are now verifiable in production. Install: brew upgrade initializ/tap/forge Full release notes → go.useforge.ai/rel-v0130

Forge v0.10.1 🩹 🪟 Windows Fix — forge ui no longer falsely reports healthy agents as failed daemons
📚 Docs Cleanup — fixed 55 broken links across README + docs
🧪 CI Protection — added automated validation to prevent future broken documentation links Small release. Important upgrade for Windows users. brew upgrade initializ/tap/forge Full release notes → go.useforge.ai/rel-v0101

Forge v0.10.0 🚀 🔐 Dynamic Secret Overlay — skill-declared secrets now flow directly into the agent runtime environment
📊 Configurable Security Scoring — custom audit policies + scoring overrides with inline audit annotations
☸️ Kubernetes Channel Env Injection — Slack/Telegram env vars now automatically wired into generated manifests
🤝 Slack Bot Admission — allow CI/scheduler bots to safely @mention agents with built-in self-loop protection
🧹 Secret Chain Recovery — stale encrypted secret stores are automatically detected and isolated Forge agents just got more production-ready for enterprise deployments. brew upgrade initializ/tap/forge
docker pull ghcr.io/initializ/forge:v0.1… Full release notes → go.useforge.ai/rel-v0100
Docs → go.useforge.ai/docs

Forge v0.8.0 🔧 🧠 Code Agent — autonomous code generation + multi-framework scaffolding 🛡️ Skill Guardrails — 4-layer security policy enforcement ☸️ Kubernetes Cost Visibility — cluster spend breakdowns 💬 Telegram reliability fixes 🔎 Smarter PII detection Agents just got safer (and more useful) go.useforge.ai/rel-v080?ref=…

Checkout github.com/initializ/forge and star the project