Give Any Application Intelligence Without Changing Its Codebase
We took Sock Shop — an 8-service microservices app built around 2016, long before AI agents existed — and added revenue analysis, fraud detection, inventory intelligence, and customer profiling. Without changing a single line of code.
Too Powerful for One Terminal
Claude Code, Codex, and similar AI agents can do incredible things individually. But try running five in parallel and it falls apart.
Without Orchestration
- 5 terminal windows, each running an agent
- Copy-pasting context between sessions
- Manually tracking which agent finished what
- Sequencing the next step by hand
- Scale to 17 agents and it's unmanageable
With AgentFlow
- Sprints handle sequencing automatically
- Parallel execution across bounded contexts
- Triggers chain sprints together
- Shared memory preserves context between agents
- Policy enforcement keeps agents governed
Orchestrate Agents at Scale. In the Cloud.
AgentFlow is the control plane for AI agents. Define work as sprints, assign agents to domains, and let the platform handle sequencing, parallelism, and coordination.
Sprint Sequencing
Queue complex work across sequential and parallel sprints. Each sprint runs its agents, validates results, and triggers the next phase automatically.
Parallel Execution
Run 17+ agents simultaneously across bounded contexts. Each agent works in isolation with its own tools, memory, and policy constraints.
Shared Memory & Triggers
Agents share context through structured memory. Triggers chain sprints, events propagate results, and the entire flow is observable and auditable.
Isolated Environments
Each agent runs in its own DevContainer — dedicated services, databases, and tools. Container boundaries guarantee isolation. HIVE manages infrastructure, AgentFlow manages cognition.
Full Auditability
Every session recorded. Every agent action logged with actor attribution. Full transcripts you can replay, analyse, and use to make your agents more efficient over time.
Agent-to-Agent Communication
Assign multiple agents to a project. Message any agent, and it can discover its teammates, delegate work, and coordinate in real time. Interrogate codebases, check coverage, or get answers — agents collaborate like a team.
Your Laptop Is the Control Tower. The Containers Are the Workforce.
Each project gets its own isolated environment with its own services, databases, tools, and AgentFlow runner. URLs not ports. Ephemeral compute, persistent code. Same architecture from laptop to team server to cloud.
- Isolated container per project with full toolchain
- HIVE manages infrastructure, AgentFlow manages cognition
- Smart resource management — idle detection, auto-stop, priority allocation
- Cross-container pipelines with target_container routing
- Agent isolation guaranteed by container boundaries
- Same architecture: laptop, team server, or cloud
AgentFlow CLI
$ af dev up hive-api paymentsCreating DevContainer "hive-api"... Image: ostec/agentflow-dev:java21 Services: MongoDB 7, Redis 7 URL: https://hive-api.dev.local Runner: connected
Creating DevContainer "payments"... Image: ostec/agentflow-dev:node20 Services: PostgreSQL 16 URL: https://payments.dev.local Runner: connected
2 environments ready. 2 runners attached.DevCloud: The Only Cloud Dev Environment with a Brain
Gitpod and Codespaces give you a container. AgentFlow DevCloud gives you a container with sprint sequencing, parallel agents, shared memory, and policy governance already inside it. Accessible from any browser, anywhere.
Give Any Application Intelligence
WITHOUT changing its codebase
B2C: Intelligent Applications
Make your application intelligent for end users. AgentFlow adds cognitive capabilities to any product without changing the underlying codebase.
- Customer intelligence, fraud detection, and personalised insights
- Orchestrate multi-step user workflows autonomously
- Trigger agent pipelines from user actions via webhooks
- Scale intelligence across your entire user base
- Zero code changes to the target application
B2B: Business Intelligence Layer
Layer business intelligence on any existing system. AgentFlow orchestrates analysis, reporting, and decision-support agents across your enterprise tools.
- Revenue analysis, inventory forecasting, and operations monitoring
- Cross-system intelligence without migration or integration projects
- Automated compliance, security, and audit workflows
- Domain-agnostic: same platform for any business function
- Connect via webhooks, Kafka, or scheduled triggers
We Took a Pre-AI Application and Made It Intelligent
Weaveworks Sock Shop — an 8-service microservices e-commerce application built around 2016. Go, Java, Node.js, and Python services, 4 databases, RabbitMQ messaging. Nearly a decade old. Built long before AI agents existed.
We pointed AgentFlow at it. 17 agents across 7 sprints analysed the running application, probed its live APIs, and built a complete cognitive intelligence layer.
Revenue Analysis
Correlated order, product, and pricing data across 3 microservices to identify revenue patterns and pricing anomalies.
Fraud Detection
Identified PCI violations, expired card processing, and built user risk profiles from live transaction data.
Inventory Intelligence
Analysed stock levels and demand patterns to surface forecasting insights without accessing any database directly.
Customer Profiling
Built customer intelligence profiles and ran 5 AI-simulated customer service interactions against the live API.
Operations Assessment
Established performance baselines, ran concurrency testing, and assessed infrastructure readiness.
The Holy Sock Paradox
The most expensive product ($99.99) plus flat shipping ($4.99) equals $104.98 — exceeding the $100 payment threshold. A product mathematically impossible to purchase. Discovered by correlating data across 3 agents — something no single tool or human review had caught.
Every Decision. Every Action. Fully Auditable.
All 26 reports, every agent session transcript, every MCP tool call — recorded and replayable. Go back through any conversation, analyse what worked, and use those insights to make the next sprint even more efficient.
“The agents change. The tools change. The targets change. The orchestration engine stays the same. That’s AgentFlow.”
Real Numbers. Real Demos. Zero Theory.
Every claim is backed by a real demonstration you can inspect. No mockups, no “coming soon” — production-quality output from production-scale orchestration.
Cognitive Intelligence
Complete intelligence layer on a live 8-service microservices app — revenue analysis, fraud detection, customer profiling — zero code changed.
Autonomous Penetration Testing
16 vulnerabilities discovered (3 critical at CVSS 9.1–9.8), 82% exploit success rate, zero human intervention from recon to report.
PRD-to-Code Assembly Line
19,000+ lines of production Java generated from a single PRD — 13 pipeline stations, 7 bounded contexts, full Spring Boot microservice.
Open Source Contributions
8 pull requests submitted to VS Code (40M+ lines), Kubernetes (4M+ lines), Grafana, and DefinitelyTyped — all autonomously.
Proven Across Every Domain
The same platform that builds intelligence layers also runs penetration tests, generates production code, and contributes to open-source projects.
No Problem Is Too Big
Cloud-scale orchestration with sprint sequencing, parallel execution, triggers, and shared memory. Trigger pipelines from webhooks, Kafka topics, cron schedules, or internal events — domain-agnostic by design.
Webhooks
Inbound webhooks from any system — GitHub, monitoring tools, trading platforms, CI/CD. HMAC-SHA256 and API key authentication built in.
Kafka & Event Streams
Subscribe to Kafka topics and event streams. Kafka Connect for database CDC, cloud service connectors, and real-time data feeds.
Cron & Scheduling
Schedule pipelines with cron expressions. Nightly security scans, weekly compliance reports, hourly monitoring — fully automated.
Event Subscriptions
Filter and match events with field-based conditions or LLM-powered reasoning. Only trigger pipelines when criteria are met.
Event-Driven Patterns
Reactive Pipeline: Webhook → Analyse → Decide → Act # Price alert triggers market analysis pipelineChain Reaction: Pipeline A → Event → Pipeline B # Build completes, triggers deployment pipelineContinuous Monitoring: Stream → Batch → Periodic Reasoning # Batch alerts every 15 minutes for trend analysisExtensible Agent Toolkit
357+ MCP tools enable domain agents to interact with their environment — from project management to deployment. The toolkit powers the orchestration, not the other way around.
"Run the full pipeline from this PRD""Decompose the auth system into bounded contexts""Execute phase 3 across all assigned agents""Show pipeline progress and agent status""Transfer this pipeline template to the pentesting domain"Orchestration Capabilities
See It in Action
Request a walkthrough of AgentFlow orchestrating 17 agents across 7 sprints on a live application — or submit your own requirements and see what comes back.