Control Plane
A design pattern borrowed from networking and cloud computing architecture that, in the context of AI agents, refers to a governance layer operating above model execution logic. It is responsible for centrally managing permission policies, enforcing approval workflows, providing sandbox isolation, and offering audit observability.
Core Facts
Timeline (last 90 days)
The author designed the system starting from one rule: the coordinating Agent's service must never be able to read code.
The control plane mints a per-run token, an HMAC signature based on the run id and expiration time, valid for 30 minutes and restricted to a single run.
The control plane starts workflows via workflow_dispatch, which is the technical fulcrum that makes the rule "the coordinator can never read code" achievable.
The control plane, as a GitHub App, has only two permissions, Metadata: read and Actions: read & write, and has no Contents permission, so it cannot clone repositories or read files.
The system consists of two components: the control plane is responsible for managing error messages, configuration, and run history; the Runner is the user's own GitHub Actions runner, which holds the code and the AI keys.
In current benchmarks, the recall plane has been extensively benchmarked, while the control plane is far from adequately covered when faced with production systems actually deployed.
The recall plane corresponds to the read path, and its metrics are usually recall, precision, and latency; the control plane corresponds to the write path and lifecycle management, including adding, superseding, revoking, expiring, and purging facts, as well as mechanisms for propagating and acknowledging expired state.
The data plane is responsible for actually forwarding each request, emphasizing low latency and high throughput; the control plane does not touch user requests themselves and is only responsible for distributing and updating configuration.
TheYgent's architectural rule is that the control plane never imports the engine, contains no MLX, vLLM or llama.cpp bindings at all, and accesses inference capability only through the inference plane's OpenAI-compatible HTTP API.
TheYgent is a local-first AI Agent platform with a two-plane architecture: the inference plane runs models (engine processes, model weights, model registry, API credentials), and the control plane orchestrates Agents (graph runtime, Agent registry, sessions, triggers, MCP servers, retrieval sources, secrets and run traces).
11 more timeline events
All Facts (20)
The layered architecture of control plane and data plane originated in network device design, where the control plane is responsible for decision-making and coordination, and the data plane is responsible for actual execution.
75%UnverifiedThe control plane mints a per-run token, an HMAC signature based on the run id and expiration time, valid for 30 minutes and restricted to a single run.
50%UnverifiedThe author designed the system starting from one rule: the coordinating Agent's service must never be able to read code.
50%UnverifiedThe system consists of two components: the control plane is responsible for managing error messages, configuration, and run history; the Runner is the user's own GitHub Actions runner, which holds the code and the AI keys.
50%UnverifiedThe control plane, as a GitHub App, has only two permissions, Metadata: read and Actions: read & write, and has no Contents permission, so it cannot clone repositories or read files.
50%UnverifiedThe control plane starts workflows via workflow_dispatch, which is the technical fulcrum that makes the rule "the coordinator can never read code" achievable.
50%UnverifiedIn current benchmarks, the recall plane has been extensively benchmarked, while the control plane is far from adequately covered when faced with production systems actually deployed.
50%UnverifiedThe recall plane corresponds to the read path, and its metrics are usually recall, precision, and latency; the control plane corresponds to the write path and lifecycle management, including adding, superseding, revoking, expiring, and purging facts, as well as mechanisms for propagating and acknowledging expired state.
50%UnverifiedThe data plane is responsible for actually forwarding each request, emphasizing low latency and high throughput; the control plane does not touch user requests themselves and is only responsible for distributing and updating configuration.
50%UnverifiedTheYgent is a local-first AI Agent platform with a two-plane architecture: the inference plane runs models (engine processes, model weights, model registry, API credentials), and the control plane orchestrates Agents (graph runtime, Agent registry, sessions, triggers, MCP servers, retrieval sources, secrets and run traces).
50%UnverifiedTheYgent's architectural rule is that the control plane never imports the engine, contains no MLX, vLLM or llama.cpp bindings at all, and accesses inference capability only through the inference plane's OpenAI-compatible HTTP API.
50%UnverifiedIn cloud service architecture, the control plane handles configuration, monitoring, billing and policy distribution, while the data plane directly carries users' business requests.
50%UnverifiedThe concept of control plane originates from network engineering, used to distinguish the data plane (responsible for forwarding packets according to rules) from the control plane (responsible for computing routing policies, maintaining topology state, and issuing forwarding tables).
50%UnverifiedModels, tools, MCP servers, skills, and Agents should be governed in a single unified place, as decentralized management is bound to result in loss of control.
50%UnverifiedThe control plane concept is borrowed from Software-Defined Networking (SDN), where the control plane is responsible for decision-making and coordination while the data plane handles actual data forwarding.
50%UnverifiedCommon bottleneck sources in the control plane include state storage read/write latency, task queue buildup, lock contention, and cascading latency from cross-service calls
50%UnverifiedAs concurrent request volume increases, the control plane must simultaneously maintain state machines for hundreds or thousands of sessions, and the pressure growth is often nonlinear
50%UnverifiedThe load model of the control plane is fundamentally different from that of the data plane and cannot simply adopt the scaling approach of ordinary web services; it requires independent capacity planning
50%UnverifiedThe control plane is a concept originating from network engineering and distributed systems, referring to the system layer responsible for managing, configuring, and coordinating data flows, as opposed to the data plane that processes data; in cloud-native architecture, Kubernetes' API Server is a typical control plane component.
50%Unverified在Kubernetes架构中，控制平面负责全局调度决策和事件响应，数据平面执行实际工作负载
50%