Rethunk.AI governs the operation of AI agents end-to-end - from the policy gate that authorises each directive, to the code-and-knowledge substrate where agents operate.
Rethunk.AI governs the operation of AI agents end-to-end. Bastion handles run-time governance - the policy gate that evaluates every operator directive, the hash-chained intent ledger that records each attributed action, and the evidence path from browser signing to edge execution. Citadel hosts the code-and-knowledge substrate where those agents operate - repositories, a knowledge-graph indexer, and a first-class agent identity layer in one self-hostable binary. Together they form a single governance story across the operational and engineering planes of AI agent deployment.
Left pane: Bastion. An operator sends a directive through the IRONLAW gate - a signed doctrine bundle - which routes to Theatre execution at the edge. Right pane: Citadel. An AI agent interacts with the namespace graph for permissions and identity, producing an audit feed. Centre seam: Citadel audit emissions flow into the Bastion attestation surface. Bottom rail: Rethunk.AI governance plane spans both products.
Run-time governance for AI agent operations
Bastion is a governed agent operations layer that connects an IRONLAW policy-as-code gate, an offline-verifiable intent ledger, and a cryptographic evidence chain into a single execution path from operator to edge. Operators issue directives through a browser-side signing surface; Bastion evaluates each directive against active doctrine, appends an attributed line to the master-intent ledger, and dispatches a signed desired-state payload to TheatreManager at the edge. Every step in that path is observable, attributable, and verifiable without a running server.
Bastion is built for regulated-enterprise environments where AI agent operations must satisfy board-level accountability requirements. The board-level question for every AI agent deployment is the same: what ran, under whose authority, and can you prove it? Bastion treats the evidence chain as a first-class design requirement so that answer is always available offline, without reconstructing it from secondary sources.
AI-native git forge and knowledge substrate
Citadel is a single self-hostable Go monolith that combines a git forge, a knowledge-graph indexer, and an agent identity layer in one governed substrate. Repositories, the graph derived from them, and the agents that act on them share the same permission model and the same audit surface - so there is no seam between source control, code intelligence, and AI agent operations. Everything is governed end to end.
Citadel is built for regulated-enterprise, sovereign, and defense-adjacent environments where running a forge on someone else's infrastructure is not an option. Air-gap and on-premises deployments are explicit design requirements, not afterthoughts - the single-binary shape and the air-gap install scripts exist because the target environment demands them. If your procurement or security posture requires a forge you control completely, Citadel is the substrate.
Bastion's IRONLAW policy gate and hash-chained intent ledger interoperate with Citadel's agent identity layer and audit feed. When an AI agent acts within Citadel - pushing code, querying the knowledge graph, or opening an issue - that action produces an attributable audit record through the MCP surface. Bastion can attest against that record as part of its reconcile-time evidence chain. The result is one governance story across the operational plane (who authorised the directive, what ran, can you prove it) and the engineering plane (what the agent touched, which repos, under which permissions). No integration seam between the two planes is required - they share the same accountability model by design.
Ready to evaluate the platform?
Book time with the team for a walkthrough aligned to your environment and governance requirements.
Was this page helpful?
