Context
KnowledgeQuery + ContextAssembly
Resolve relevant knowledge, world context and constraints before assembling what the provider sees.
Agent-ready
Bring your agent. Keep your business authority.
Connect OpenClaw, another agent framework or your own automation through cell, Tooling endpoints or MCP. Agents can reason, call capabilities and propose actions while application semantics, policy and execution remain owned by logiCells.
Three access paths
Use the agent stack you already have.
Agent access is a surface of the application, not a second business model. The same concepts, capabilities and runtime rules remain authoritative regardless of the caller.
CLI
cell agent
Run goals, open persistent sessions, inspect pending confirmations, approve or reject requests and evaluate sessions from the command line.
Tooling
Persistent runtime endpoint
Let an external orchestration layer work against the application already loaded in the runtime instead of starting a second copy of it.
MCP
Governed external capabilities
Discover and call MCP capabilities through the same catalog, policy and confirmation boundaries used by native application actions.
Native operationalization
MCP is a backend, not the business identity.
A semantic Action remains stable even when its execution backend changes. MCP capabilities can be operationalized through native AST, so protocol details stay behind the application contract instead of becoming the contract.
01Semantic Action
02#has_operational → #ast_program
03native AST wrapper
04MCP.Call / native / service
05typed result · EvidenceSet · provenance
Governed execution
The agent proposes. The runtime decides what can execute.
The LLM is not the business authority. Capabilities are typed, policy-aware and can require explicit confirmation before an effect is dispatched.
Catalog + policy
Discovery is not authorization.
A capability can be visible to the system without being automatically permitted for the current agent, user, scope or context.
Typed binding
Arguments are validated before execution.
The runtime binds agent-supplied values to native action contracts rather than trusting arbitrary protocol payloads.
Confirmation
Effectful operations can stop before the effect.
Pending confirmations create an explicit boundary where approval or rejection is owned by the host, not by the model.
Native execution
Approved actions use the application runtime.
The final operation resolves through the same action and runtime mechanisms used by the application itself.
Knowledge Runtime
Give agents structured context, evidence and scoped knowledge.
Agent context can be assembled from conceptual knowledge instead of relying only on retrieval over unstructured text. Scope, provenance, evidence and policy remain visible through the query and context pipeline.
Evidence
BindingSet + EvidenceSet
Carry semantic results together with the evidence and provenance needed to understand where they came from.
Shared knowledge
Personal → team → project → organization
Promote knowledge between explicit scopes through governance rather than flattening personal memory into a single company truth.
External agents
OpenClaw is one caller, not a dependency.
An external agent framework can drive logiCells through CLI, Tooling or MCP while the application retains its own semantic actions and governance. If the framework or model changes later, the business capability does not need to be renamed around the integration.
Agent-ready Community
Build agents on top of executable business meaning.
Use agents where they add leverage, while keeping the application model, knowledge boundaries and execution rules explicit.