/
/
Cogentiq
Agents cannot access the systems your decisions depend on.
Risk and compliance teams cannot verify that agents are safe to run.
Your second use case costs as much to build as your first.
Reach the information your business runs on
Give agents access to data across your existing systems. Federated reasoning reads that data where it lives.
Carry your controls into every run
Apply shared policies through one control plane, so safeguards carry across agents and use cases.
Reduce the cost of your next agent
Reuse a shared library of agents, tools, and business context as you add use cases.
1
CONTEXT
What your business means
Ontologies, glossaries, data bindings, tools, prompts, and skills in one graph
2
CONNECT
Where the data actually is
Secure connectors, retrieval, and an MCP gateway over structured and unstructured data
3
BUILD
Agents that use both
Low-code canvas for domain experts and a pro-code SDK on one object model
4
RUN
Orchestrated, with people in the loop
Multi-agent orchestration, stateful execution, and approval gates by named role
5
GOVERN
Policy, cost and audit in one place
RBAC, SSO, guardrails, evaluations and audit trails on every agent by default
Establish industry knowledge once and reuse it across every solution.
Set your definitions, policies, structures, and thresholds once.
Extend that shared foundation with context specific to the new solution.
Our partners
Our expert
Abhijit Guha
Client Partner
Cogentiq Platform
Platform basics
What is Cogentiq?
Cogentiq is Fractal’s enterprise agentic AI platform. It brings together agent-building capabilities, a context and connectivity layer for existing systems, orchestration for teams of agents, and a unified control plane for governing agents in production.
How is Cogentiq different from building on an LLM API or a cloud agent framework?
LLM APIs provide access to models, while cloud frameworks help teams build with them. Cogentiq adds the enterprise layer needed to operationalize agents, including system connectors, shared context, approval gates, evaluations, cost attribution, and auditability, while allowing organizations to choose the models that best fit their needs.
Which models and clouds are supported?
Cogentiq is designed to be model- and cloud-agnostic. Different agents, and even different steps within an agent, can use different models. Model selection is managed through configuration rather than hard-coded into the agent, helping teams adopt new options without rebuilding the full solution.
Who can build agents: business teams or engineers?
Both can contribute using a shared object model. Domain experts can assemble agents through a low-code canvas, while engineering teams can extend the same agents in code when deeper customization is required.
How does Cogentiq help prevent citizen development from becoming shadow AI?
Agents created through the canvas inherit platform policies such as role-based access, guardrails, evaluations, control-room visibility, and review before publication. This enables broader participation in building agents while keeping production use governed.
Data, security, and control
Do we have to move our data?
Not necessarily. Cogentiq can use secure connectors to reason across systems where data already resides, while retrieval grounds responses in existing enterprise knowledge. Access remains aligned with permissions enforced by the source systems.
How are responses grounded, and what happens when an agent is wrong?
Responses can be grounded in retrieved enterprise content and presented with source references. Evaluations can assess qualities such as groundedness, while guardrails can limit out-of-scope requests. Run traces help teams review which agent acted, what data was used, and where an issue occurred, supporting investigation and improvement.
Can people remain involved in high-stakes decisions?
Yes. Human review can be built directly into the workflow through approval gates. A run can pause until an authorized role provides sign-off, then continue from that point without restarting the entire process.
How is access controlled, and can data remain within a region?
Cogentiq supports single sign-on and role-based access control scoped by agent. Regional residency and retention requirements can be defined through platform policies and applied consistently across agents, subject to the selected deployment and service configuration.
Can Cogentiq run inside our cloud environment?
Cogentiq supports multi-cloud deployment and can be deployed in customer-managed environments. Private networking, restricted egress, and model-endpoint requirements should be considered early because they can shape the final architecture.
What compliance evidence is available?
Fractal can provide relevant certification reports, penetration-test summaries, and a completed security questionnaire under NDA. Bringing the evidence review into the evaluation phase can help security assessment progress alongside the pilot.
Commercials and rollout
How long does it take to put a first agent into production?
For a clearly scoped process with a named data owner and business sponsor, a six-to-eight-week plan may be achievable. Timelines depend on factors such as access approvals, baseline agreement, integration complexity, and sign-off authority.
How is Cogentiq priced?
Cogentiq is offered through a platform subscription. Commercial terms can be aligned to the scope, deployment model, and enterprise requirements.
Can we buy Cogentiq through AWS Marketplace?
Cogentiq is available through AWS Marketplace in the AI Agents and Tools category, which may allow eligible organizations to purchase through an existing committed-spend agreement. It is also available directly from Fractal.
What does Fractal provide beyond the software?
Fractal combines the Cogentiq platform with AI, engineering, and domain expertise. For early use cases, this can help organizations move from process definition and context design through implementation and adoption.
How should we measure return on investment?
Start with the business process rather than the model. Establish a baseline before go-live, use a consistent measurement window, and include the full operating cost of the agent, including inference and human-review effort. The final measures should reflect the outcomes the use case is intended to improve.
How does Cogentiq work with automation tools we already use?
Cogentiq is designed to complement existing automation. High-volume, rules-based steps can remain in current tools, while Cogentiq can support work that requires judgment across complex or unstructured inputs and can call existing systems as tools within a run.
Cogentiq Context
What is Cogentiq Context?
Cogentiq Context provides a shared business-context layer for agents. It brings together elements such as ontologies, glossaries, data bindings, tools, prompts, and skills so agents can work with enterprise-specific meaning rather than relying only on the data presented in a single interaction.
Why does business context matter, even when using a more capable model?
A model understands language, but it does not automatically know how your organization defines terms, applies policies, or determines which source is authoritative. Shared context helps agents use the definitions and rules that matter to the business, reducing the risk of confident but misaligned responses.
What are the three dimensions of context?
Domain context captures what is true for an industry. Organizational context reflects company-specific policies, definitions, and ways of working. Solution context covers what is specific to an individual use case. Domain and organizational context can be reused as additional agents are introduced.
Do we need a complete ontology before getting started?
No. A practical approach is to model the entities, terms, and bindings required for the first use case, then expand the context graph as new agents and use cases are added.
Can we reuse our existing glossary or data catalogue?
In many cases, yes. Existing business glossaries, data catalogues, and semantic layers can serve as inputs to the context graph rather than being replaced. The structure and readiness of those assets can influence how quickly bindings are established.
Who owns and maintains the context graph?
Ownership is typically shared. Business teams steward definitions, data engineering manages bindings, and security teams govern tool permissions. Named stewards for each domain help keep the context current and consistent as it evolves.
How does context relate to retrieval and RAG?
Retrieval identifies content that appears relevant. Context helps an agent interpret what that content means within the business and determine which sources and definitions should take precedence. Used together, they can produce responses that are more relevant and aligned to enterprise requirements.
How long does it take to build a useful context graph?
A first use case can begin with a focused set of terms, entities, and bindings. The timeline will depend less on volume than on the availability of source systems and agreement across teams on definitions that may be interpreted differently.
