All I did was
ASK a QUESTION...
...getting answers can be a challenge.
Every request that goes unanswered costs money.
“How much do I owe you?”
Every unanswered inquiry incurs storage costs in one or more systems -- for every participant in an exchange.
As precious time goes by with no progress, information is lost and participants disengage:
“How did you find out about us?”
“I don't remember, I've been a customer for 20 years”
Streaming chatbots serving as front-end interfaces for Retrieval-augmented generation (RAG) systems DO answer questions, BUT:
- Only SOME OF THEM.
(They don't know who I am, and if they did, they shouldn't retrieve my customer information unless they can identify me as the authorized consumer.) - Setting a meaningful context for more nuanced questions can be problematic.
(Long explanations about the actual question often introduce tangents that increase processing costs.) - When a question is a request, as in, “I want you to do something for me.”
(The question is actually a Request in a Request-Response pair, correlated with a Message-Id, AND
a Task, correlated with a Task-Id. ) - When a request is "long-running", or asynchronous, responses could be orphaned and information is lost.
(A Task-Id associated with a Request using this Message-Id that has no Response is a Task with NO STATUS.) - Yo-AI correlates Request-Response pairs for Messages and Tasks to insure that information is never lost.
(Orphaned messages go into the Dead-Letter-Queue)
WHO Is
ASKING?
In Yo-AI, before a Request (Message or Task) is routed, the most important context is WHO THE SENDER IS.
Another agent, the Door-Keeper, intercepts the Request and promotes the Sender's identifying information to YoAiContext, which sets the Actor properties:
class YoAiContext(TypedDict, total=False):
"""
The single unified interchange context for the YoAi platform.
Constructed once per request-response interchange and passed as a call
parameter into every capability handler. Never stored on agent instances.
Two temporal faces
──────────────────
REQUEST face Populated at construction from the incoming envelope.
Answers: who is asking, on whose behalf, for what, how.
RESPONSE face Written by the capability as it produces its result.
Answers: what changed, what should travel forward.
┌─ Correlation ──────────────────────────────────────────────────────┐
│ correlation_id JSON-RPC id. Primary request-response handle. │
│ Set by A2ATransport at the protocol boundary. │
│ task_id A2A task identifier. │
│ Defaults to correlation_id if absent. │
└────────────────────────────────────────────────────────────────────┘
┌─ Actor (REQUEST) ──────────────────────────────────────────────────┐
│ actor_kind What kind of entity invoked this operation. │
│ actor Identity dict of the invoking entity. │
└────────────────────────────────────────────────────────────────────┘
┌─ Invocation mechanics (REQUEST) ───────────────────────────────────┐
│ startup_mode How the request entered the platform. │
│ instance_id Runtime identity of the handling agent instance. │
│ None for PlatformAgents. │
│ caller Registered caller identity for trust-gated access.│
└────────────────────────────────────────────────────────────────────┘
┌─ Subject (REQUEST) ────────────────────────────────────────────────┐
│ profile ProfileWrapper — lightweight profile snippet. │
│ Shape: {type, name?, payload} │
│ subject_ref Lightweight pointer to the request subject. │
└────────────────────────────────────────────────────────────────────┘
┌─ Capability identity (REQUEST) ────────────────────────────────────┐
│ capability_id The capability being invoked, e.g. "Trust.Assign"│
│ None at pipeline level; bound via │
│ ctx_for_capability(). │
└────────────────────────────────────────────────────────────────────┘
┌─ Execution knobs (REQUEST) ────────────────────────────────────────┐
│ slim Skip expensive init (fingerprints, knowledge, │
│ tools). │
│ tools Selective tool loading. │
│ None=all, []=none, ["vault"]=named subset. │
│ dry_run Validate without executing side effects. │
│ trace Activate OTel Layer 4 explainability tracing. │
└────────────────────────────────────────────────────────────────────┘
┌─ Workflow state (REQUEST, updated per step) ───────────────────────┐
│ step Current workflow step index. │
│ prior_outputs Outputs from previous steps. │
│ state Arbitrary workflow state bag. │
└────────────────────────────────────────────────────────────────────┘
┌─ RESPONSE face — written by the capability, travels forward ───────┐
│ profile_patch Fields discovered during execution not present in │
│ the original profile. Forwarded to Data-Steward. │
│ Shape: {type, payload} │
│ governance_labels Sticky notes attached by the capability. │
│ NOT present in request envelopes — prevents │
│ information spills between capabilities. │
│ Shape: list[str] │
└────────────────────────────────────────────────────────────────────┘
"""
# ── Correlation ────────────────────────────────────────────────────
correlation_id: str | None
task_id: str | None
# ── Actor ──────────────────────────────────────────────────────────
actor_kind: ActorKind | None
actor: dict[str, Any] | None
# ── Invocation mechanics ───────────────────────────────────────────
startup_mode: StartupMode | None
instance_id: str | None
caller: dict[str, Any] | None
# ── Subject ────────────────────────────────────────────────────────
profile: dict[str, Any] | None # ProfileWrapper
subject_ref: dict[str, Any] | None
# ── Capability identity ────────────────────────────────────────────
capability_id: str | None
# ── Execution knobs ────────────────────────────────────────────────
slim: bool
tools: list[str] | None
dry_run: bool
trace: bool
# ── Workflow state ─────────────────────────────────────────────────
step: int | None
prior_outputs: dict[str, Any]
state: dict[str, Any]
# ── RESPONSE face ──────────────────────────────────────────────────
profile_patch: dict[str, Any] | None # ProfilePatch
governance_labels: list[str]
YoAiContext provides metadata to support custom processing operations, setting tags in the governanceLabels of the payload envelope for use in downstream processes for persistence.
GET
Yo-AI
Discover what YOU ARE MISSING.