BAMS, the BlackUnicorn Agentic Management System, is the platform we run our own agent fleet on, delivered as a custom Enterprise Edition installation inside infrastructure you control. Your models, your knowledge, your approval paths, and a handover that leaves your team able to operate it without us.
A capable model can generate an answer. An enterprise AI system also needs to know which agent is acting, what that agent may access, which model may receive the request, which organizational knowledge applies, and when a human decision is required.
It needs to know how work moves from intent to evidence, how to stop a failing agent or provider, and what happened, who approved it, and what it cost. Stitch frameworks and APIs together without deciding those questions, then bolt governance on after launch, and it gets expensive and fragile.
BAMS puts those decisions in the control plane. It is not a chat wrapper and it is not a collection of disconnected agents.
The operator sees current work, decision queues, fleet health, risk signals, and infrastructure state without reconstructing the story from model logs and vendor consoles.
Projects carry a charter, ownership, classification, members, milestones, tasks, activity, and journals. Governance rules are enforced in the workflow: charter before active, RACI completeness, weekly journals, review before final completion. In the autonomous work path, an agent reporting completion proposes a review state. It does not silently mark its own work final.
Local and hosted model routes, OpenAI-compatible providers, per-agent chains, provider health, quotas, and Vault-backed credentials. BlackUnicorn is EU-domiciled, and the stack deploys inside your boundary: no required BlackUnicorn cloud and no forced model provider.
BoundaryBAMS does not pretend that every model weight is yours. It gives you control over the models and providers you are entitled to run, and it keeps the routing decision visible.
Memory is scoped globally, per agent, and per session, with cross-read ACLs, operator inspection and correction, and unified search. Documents, credentials, and evidence stay inside the topology you approve.
BoundaryScoped memory governs what agents recall inside BAMS; it does not retrofit governance onto data you send to external providers outside its routing.
An agent in BAMS has a declared identity, organizational unit, leader, lifecycle state, skill set, access policy, model route, memory scope, schedule, telemetry, and audit lineage. The internal reference deployment runs a 25-agent main fleet, and BAMS also supports isolated protected tiers for sensitive work. The isolation function is the public product claim; protected names, missions, personas, hosts and tier sizes are not exposed.
BAMS uses three static autonomy tiers. Lower-risk work proceeds under policy. Higher-risk work creates a typed decision carrying request context, urgency, routing, status, and decision history. Communication and high-impact actions can be gated by approval and allowlist policy instead of relying on a prompt to behave.
BoundaryThe tiers are static and policy-defined. A per-decision risk classifier exists in preview and is disabled by default, so BAMS does not claim automatic risk escalation.
A per-agent Control Debt score is derived from operational evidence, and when control quality drops below threshold it auto-trips a circuit breaker. Autonomy is earned by evidence and withdrawn by the same mechanism.
BoundaryControl Debt trips the breaker; it does not diagnose why control decayed. That stays an operator investigation with the audit trail as the receipts.
CB-1 halts the platform. CB-2 halts a specific team or agent. Both are effectful controls exercised in fire drills. CB-3 through CB-5 record degraded operation, recovery, and post-incident state so the story survives the incident.
The governance surface is aligned to an AI management system and mapped to ISO/IEC 42001 operating concerns: governed-agent inventory, policy catalog, RACI, data classification, risk register, KPIs, and evidence, plus runtime measures, Control Debt, a validation ratio quantifying human oversight, and an HMAC-chained audit trail with integrity verification.
The aim is not to claim certification from a dashboard.
The aim is to make controls inspectable and give assurance work concrete receipts. Mapped to ISO/IEC 42001 operating concerns is the claim; certified is not.
Only CB-1 and CB-2 are effectful runtime controls today.
CB-3 through CB-5 are recorded governance states for degraded operation, recovery, and post-incident handling. Fire drills exercise the effectful pair.
Connector readiness varies by deployment.
The integration board distinguishes connected, degraded, and unconfigured services instead of implying every connector is ready everywhere.
The infrastructure surfaces cover deep health, model capacity, memory services, provider status, storage tiers, Vault connectivity, service metrics, and fleet readiness. Operators can see whether an agent failure is a reasoning problem, a policy stop, an exhausted provider, or an unhealthy dependency.
A BAMS engagement is a custom, services-led Enterprise Edition installation: deployment engineered around your security boundary, integration of approved knowledge sources, models, and providers, policy and approval configuration, acceptance evidence, operating runbooks, circuit-breaker acceptance testing, and a forward-deployed engineer through handover. The acceptance criterion is written into the PRD: your team can perform the agreed normal operating procedures without a BlackUnicorn operator.
No strings attached does not mean no license.
BAMS remains proprietary and commercially licensed. It means no required BlackUnicorn cloud, no forced model provider, and no requirement to retain BlackUnicorn for day-to-day operation.
It is not a shared self-serve SaaS.
BAMS Enterprise Edition is a private, client-controlled installation. The standardized appliance installer and self-service onboarding are separate roadmap work.
BAMS is designed for organizations that require control over deployment infrastructure and credentials, model and provider selection, organizational knowledge and retained context, agent authority, approval and escalation paths, security controls and runtime recovery, and evidence for operational review. If none of that list matters to you, a hosted agent product is probably the better buy.
A deployment conversation. Each installation is engineered around your security boundary, organization, knowledge sources, model policy, integrations, approval paths, and operational requirements, so the scoping call is the real first step. Leave your email below and we reply from a human inbox.
There is no published price. BAMS is a custom, services-led enterprise installation, and the scoping call is where a number becomes honest.
The platform BAMS ports is the one we run our own operation on. The screenshots on this page are its current interface running built-in demo fixtures, labelled SIMULATED DATA, because our production data is not marketing material.
Proprietary, commercially licensed, deployed privately inside your boundary. The open-source parts of the BlackUnicorn stack (DojoLM, BonkLM, RuneLM) are separate products with their own licences.
Tell us about your deployment and we will be in touch. Reply inside 48 hours, scoped within a week, NDA-first.
BAMS is the platform we run our own fleet on, delivered as a services-led installation inside your boundary. Talk to us about a deployment.