Authority
Identity, purpose, action tiers, approval, and evidence for agents that can cause side effects.
Founded 2024
Neul Labs builds the systems around the model.
I founded Neul Labs to work on the layer that determines whether an AI system can be trusted with real work: state, authority, evaluation, execution boundaries, recovery, and operational ownership.
The thesis
The model is one component of the operating system
When an AI system can call tools, write code, change data, or affect the physical world, reliability depends on the architecture around inference.
Durability
Explicit state, checkpoints, idempotent effects, replay, reconciliation, and recovery.
Evaluation
Exact and adversarial tests that reveal plausible but incorrect code or agent behaviour.
Inference
Local, embedded, and multi-provider execution with explicit resource and compatibility boundaries.
Containment
Custom sandboxes, isolated workspaces, scoped credentials, and clear tool interfaces.
Operations
Observability, budgets, versioning, kill switches, rollback, and an accountable human owner.
Public systems
Runtime governance, durable execution, and agent coordination
Regulus
Runtime governance for identity, purpose, policy, privacy, model risk, oversight, and evidence.
View source ↗fast-langgraph
Durable agent state and checkpoint operations, with carefully scoped project benchmarks.
View source ↗grite
Git-native coordination events for measuring multi-agent work before the pull request.
View source ↗agentvfs
A public execution-boundary reference for isolated agent workspaces.
View source ↗brat
A crash-safe multi-agent harness and practical coordination artefact.
View source ↗Neul Labs GitHub
The current public organisation index and source for project status.
View source ↗The linked repositories contain source, documentation, and measured project results. No external audit, certification, or customer deployment is published for these systems.
Direct answers
About Neul Labs
Who founded Neul Labs?+
Dipankar Sarkar founded Neul Labs and works through Neul Labs Limited as a principal consultant and hands-on applied AI systems engineer.
What does Neul Labs work on?+
Neul Labs works on dependable AI and agent infrastructure: runtime governance, durable orchestration and recovery, model routing, evaluation, local inference, execution boundaries, and operational tooling.
Is Neul Labs work open source?+
A substantial body of public reference work is available through the neul-labs GitHub organisation. Public code establishes inspectable design and implementation work; it does not imply customer adoption or certification.
Where is Neul Labs based?+
Dipankar Sarkar is based in St Andrews, Scotland, and normally delivers work remotely through Neul Labs Limited.