Command a fleet of agents like a workforce.
Mockingjay is where single agents become an organisation. Supervisors decompose objectives, worker fleets execute in parallel across your systems, shared memory keeps them coherent, and every decision stays inside policy — for days or months at a time.
- Collector100%
- Reconciler78%
- Drafter46%
- Reviewer18%
Concurrent agents coordinated inside a single governed fleet.
Reduction in cycle time on multi-team, multi-system processes.
Of agent decisions captured in a replayable audit trail.
Process duration — fleets survive restarts, upgrades and handovers.
What Mockingjay does
Agentic AI with orchestrated agent fleets.
Agent fleets
Define roles, team structures and escalation paths. Scale a fleet from three agents to three hundred without redesigning it.
Shared fleet memory
A common working memory keeps every agent aligned on facts, decisions and open questions across the whole run.
Durable long-running processes
Checkpointed state means a process that takes six weeks survives failures, deployments and shift changes.
Policy & guardrail engine
Declarative rules on what each role may see, spend, sign or send — enforced before the action, not audited after.
Simulation & replay
Rehearse a fleet against historical scenarios, then replay any real run step by step to see exactly what happened.
Workforce analytics
Throughput, handoffs, bottlenecks and cost per outcome — the operating metrics of a digital workforce.
Supervisors plan. Fleets execute. Nothing drifts.
A supervisor agent decomposes an objective into a dependency graph, assigns work to specialised roles, and reconciles their output. Conflicts route to arbitration rather than silently overwriting each other.
- Hierarchical supervisor / worker / reviewer topologies
- Dependency-aware scheduling with parallel branches
- Deterministic arbitration when agents disagree
- Backpressure and budget control at the fleet level
- Collector100%
- Reconciler78%
- Drafter46%
- Reviewer18%
Autonomy with a hand on the brake
Every fleet runs inside an authority model you define. Low-risk steps proceed; anything that crosses a threshold pauses for a named human owner, with full context attached.
- Role-scoped permissions, spend limits and approval thresholds
- Segregation of duties enforced across agent roles
- Kill-switch and quarantine for an individual agent or a whole fleet
- Evidence packs exportable for audit and regulatory review
- Read customer PIICollector roleallow
- Spend above £5,000Drafter roleapprove
- Send external emailAll rolesapprove
- Modify ledger entriesAll rolesdeny
Paused · awaiting owner
Drafter · £6,480 payment to supplier 88-214
From one pilot to an enterprise-wide digital workforce
Fleets are versioned artefacts. Prove one in a single department, then replicate it across regions and business units with local policy overrides instead of a rebuild.
- Fleet templates with environment-specific configuration
- Multi-region deployment with data-residency pinning
- Shared tool and connector catalogue across every fleet
- Capacity planning against Soika Stack GPU inventory
- eu-westRunningtemplate v4.26 fleetsresidency pinned
- me-centralRunningtemplate v4.24 fleetsresidency pinned
- apac-southRolling outtemplate v4.13 fleetsresidency pinned
Specifications
- Topologies
- Supervisor/worker, peer review, pipeline, market-based
- State
- Durable checkpointing, event sourcing, replay from any step
- Memory
- Episodic, semantic and shared working memory per fleet
- Integration
- MCP tool servers, enterprise connectors, custom APIs
- Licensing
- Annual enterprise subscription, per-fleet capacity tiers
- Deployment
- Private cloud, on-premise, air-gapped, sovereign
Where it earns its place
Regulatory reporting
A fleet gathers, reconciles, drafts and reviews submissions across dozens of source systems each cycle.
Large-scale migration
Hundreds of agents analyse, convert and verify legacy estates in parallel with human checkpoints.
Procurement to contract
Sourcing, evaluation, negotiation support and contract assembly run as one governed, auditable process.
The other two products
Every Soika product is bought, deployed and run on its own. Soika Stack is the exception in one direction only: it can serve the models behind either platform.
See Mockingjay on your own data
We run a scoped discovery workshop, deploy into your environment and prove the outcome against your real workload — with your local Soika partner alongside.