The coordination layer in supply chains is moving from people to systems.
I experiment with AI and build tools to understand what problems it can solve and where it hits its own limitations. From chasing to orchestration - till the last workflow...
Automatiqa Lab builds open-source agentic AI tools for supply chain operations, from document ingestion and risk sensing through forecasting and compliance checks to execution orchestration.
I'm Aleks Sidorecs. I've spent two decades in supply chain, where information and priorities flowed through people. Each project here takes over a piece of that coordination, and every one is built in public on GitHub.
How an agent takes over coordination
Drag sideways to read the whole schematic →
General notes
- 1Inputs: market signals (customer orders, demand forecasts, supplier lead times, commodity prices) and disruptions (freight, fuel, strikes, weather, sanctions, port congestion). The Risk Monitor watches the disruption feeds; Auspex forecasts with a confidence interval on every prediction.
- 2The agent runs the OODAA loop: observe, orient, decide, act, adjust. Rules decide wherever rules exist; a model drafts where they don't.
- 3Gate (G): a named person approves the action, as in Ship-Doc and Procezio, until the agent earns autonomy for that action type.
- 4Evals (EV) score each decision. Measured accuracy sets the autonomy level, the design of Orchestriq Core.
- 5Inventory is measured where it sits: the level transmitter (LT) on the silo sends stock positions back to observe, so every cycle starts from physical stock.
- 6Adjust: service level, cost and cash are held within control limits, and the adjust step turns each outcome into a policy update, so the loop improves across runs as well as within one. See OODAA.
Legend
- Signal or action
- Measurement
- System boundary
- Learning loop
- Approval gate
- Instrument
- Silo, stock level
- Line hop, no join
Open-source agentic AI projects for supply chain operations
-
Global Operations Risk Monitor
python agents litellm playwright fastapi MIT source ↗
Operations Risk Monitor is a multi-agent framework that watches operational risk across freight, fuel, labour, weather and geopolitics, then turns what it finds into briefings you can actually use. Seven specialist agents collect and filter the feeds into one weekly briefing.
live
-
Gatewright - open-source AI project controls for capital programmes
project controls capital projects stage gate PMO MCP MIT source ↗
Gatewright is an open-source PMO harness for programmes that run on gates - capital projects, integrations, transformations, product programmes. It scores each gate against a configurable gate model, runs chasers and forecasts, and exposes everything as MCP tools.
live
-
Ship-Doc is shipment document control for physical trade: the pack checked against the shipping instruction, drafts approved by a named person, originals chased via Outlook. The model reads, fixed rules decide, a person approves.
live
-
PPWR Compliance Screener
supply chain compliance packaging n8n rule-engine agents MIT source ↗
PPWR Compliance Screener: checks supplier packaging declarations against Regulation (EU) 2025/40 for completeness and internal consistency. Deterministic verdicts. Suppliers get a signed link and a rule engine screens what comes back; no model issues a pass or a fail.
live
-
Procezio - Process Navigator
canvas agents methodology supply chain Apache-2.0 source ↗
Procezio is a guided canvas for finding automation opportunities in the way you already work, with an AI agent working alongside you on the same board. Map the process, mark the friction, assess it, decide what to automate - the agent drafts in pencil and a human accepts.
live
-
OODAA is a small, readable self-improving agent loop. Observe, Orient, Decide, Act, and the second A, Adjust: the phase that makes an agent get smarter across runs instead of only faster within one.
live
-
Sample Trail
supply chain agri-food quality-control n8n rule-engine agents MIT source ↗
Sample Trail is a sample management workflow for the agri-food and soft-commodities sectors - coffee, cocoa, grain, sugar, cotton. It runs the quality approval decision on a physical sample, from dispatch to verdict, on the phone the person is already holding.
live
-
FlowTwin turns a recording of any operational process into editable process maps and runbooks that are adjustable, auditable, and kept current. The capture layer for tribal knowledge. Watch the process once; FlowTwin captures, segments, structures and publishes what it sees.
in development 70%
-
Supply Chain Operations Planning & Execution in Commodities (SCOPEC)
supply chain commodities S&OP S&OE planning Apache-2.0 source ↗
SCOPEC is a supply chain operations planning and execution tool for commodities: one planning kernel runs S&OP and S&OE on the same data model, commodity families live as data profiles, and agents act on exceptions under an autonomy ladder they earn.
in development 20%
-
Orchestriq Core
kernel agents evals LLMOps MCP MIT source ↗
Orchestriq Core is the kernel for agents that run real operations. Evals, approvals, earned autonomy and EU AI Act transparency live in the runtime, so every module inherits them. Spec, gate, park, earn - autonomy is a ladder climbed per action type on measured accuracy.
in development 15%
-
Auspex is a prediction and forecasting harness for supply chain operations. Connect a table, pick a model, read the signs. Zero-shot tabular prediction and time-series forecasting behind one interface, with confidence intervals on every prediction, exposed as an MCP server.
in development 10%
Supply chain heuristics, on the author's site
Asymmetric supply chain physicsSymmetric vs asymmetric supply chains, and the Outside Variance Ratio that tells you which one you're in. | heuristic 02 | ||
Aleks SidorecsRe-thinking operating models for the moment operations meet the algorithm. Who is writing, and why. | the author |