Tip: select any text on this page to bring up the Ask tooltip and question the Synedrix OS assistant about it.
What Observability is
Observability makes execution measurable, explainable, and reviewable (for every workflow, every agent invocation, every tool call, and every memory operation) without deciding whether an action was permitted or owning the business state it touched.
Three non-overlapping platforms realise this layer because "is this healthy" is genuinely three different questions with three different failure modes. Langfuse answers the model-level question, PostHog the behavioural question, the Grafana LGTM Stack the infrastructure question, with all three consuming the same canonical metric definitions and event envelope, so one correlation identifier traces across all three.
A distinction that must not be blurred: a trace is not an audit trail. Observability data is sampled and retained on an operational schedule; an audit trail must be complete and durable. That record is produced by the strictly read-only Audit & Lineage Agent, reconstructing lineage over Neo4j and Supabase rather than sampled trace data.
What it owns, and what it does not
Every layer in the architecture has a non-overlapping responsibility. The boundary is as much a part of the design as the capability.
Owns
- Logs: collected through Alloy, stored in Loki
- Traces: LLM/agent spans in Langfuse; infrastructure traces in Tempo
- Metrics: eleven canonical metric families spanning retrieval quality, hallucination, cost, tool reliability, and the memory control plane
- Run health: infrastructure and service dashboards in Grafana
- Decision lineage: read-only reconstruction of execution traces, invocation records, and policy decisions
- Alerts: threshold breaches routed with the same correlation identifiers as the triggering event
Does not own
- Whether a specific action is permitted
- → Governance & Controls Layer
- Domain state ownership: the canonical business record
- → Memory & Data Layer (Supabase PG17)
- Workflow sequencing, retries, and approval routing
- → Orchestration Layer
- Domain KPI vocabulary and targets
- → Domain Pack and Business Instance Layers
Interfaces with
The other layers this one exchanges work, state, or decisions with, and why.
Human Interface & Goal/Intent
human decisions and goal decompositions emit lifecycle events that PostHog reads to answer adoption and funnel questions.
Orchestration & Agent Runtime
every n8n workflow run and every LangGraph invocation emits traces over OpenTelemetry, the shared wire format keeping one plane coherent across the Layer 3 / Layer 4 split.
Memory & Data
the memory control plane is measured with the same metric families as any other invocation: recall mediation, consolidation lag, promotion and erasure rates.
Governance & Controls
policy decisions and tool-authorization events are emitted here so Governance can consume canonical metrics as policy inputs: a feedback loop, not a one-way street.
How Synedrix OS implements it
The Observability Metrics Spec standardises four things: a canonical event envelope with trace and subject blocks so correlation is structural rather than conventional; eleven metric families governed by six normative principles; a persistence model across six audit-grade tables in Supabase; and a platform/storage tiering rule: canonical tables in Supabase, rolling trend series in QuestDB, raw trace spans in Langfuse, alerting and dashboards through Grafana against Mimir.
Each platform then realises one slice: Langfuse captures prompt versioning, trace spans, and cost/token attribution over OpenTelemetry, unifying both n8n and LangGraph invocations into one correlated trace view. PostHog turns lifecycle events into behavioural analysis and carries the feature-flag mechanism for progressive rollout. The Grafana LGTM Stack is the operational-monitoring surface, with Alloy as the single collector forwarding to Loki, Tempo, and Mimir.
Platform mapping
Langfuse
LLM/agent trace capture, prompt versioning, cost attribution, evaluations.
PostHog
Workflow and agent behavioural analytics, feature flags, experiments.
Grafana LGTM Stack
Infrastructure health, service dashboards, alerting, and SLOs.
OS agents at this layer
Analytics Agent
intelligence · os-agt-005Aggregates cross-domain metrics and surfaces performance signals.
Forecasting Agent
intelligence · os-agt-006Generates forward projections from time-series data.
Anomaly Detection Agent
intelligence · os-agt-007Monitors operational signals and emits anomaly events for triage.
Audit & Lineage Agent
observability · os-agt-008Reconstructs what happened, why, and under whose authority. Strictly read-only.
Layer principles that apply
- Shared before specific: canonical metric families, event envelope, and correlation identifiers are a shared foundation every domain pack consumes rather than reimplementing.
- Controls before execution: policy decisions and tool-authorization outcomes are themselves observable events, not side channels invisible to this layer.
Observability questions, answered
Short, citable answers drawn from this page.
What is the Observability layer in Synedrix OS?
Observability is layer 10 of the twelve-layer Synedrix OS architecture, in the "Governance and observability" group. What happened, how well, and can it be reconstructed, for every layer, not a sample.
What does the Observability layer own?
The Observability layer owns: Logs: collected through Alloy, stored in Loki; Traces: LLM/agent spans in Langfuse; infrastructure traces in Tempo; Metrics: eleven canonical metric families spanning retrieval quality, hallucination, cost, tool reliability, and the memory control plane; Run health: infrastructure and service dashboards in Grafana; Decision lineage: read-only reconstruction of execution traces, invocation records, and policy decisions; Alerts: threshold breaches routed with the same correlation identifiers as the triggering event.
What does the Observability layer not own?
The Observability layer does not own: Whether a specific action is permitted (owned by Governance & Controls Layer); Domain state ownership: the canonical business record (owned by Memory & Data Layer (Supabase PG17)); Workflow sequencing, retries, and approval routing (owned by Orchestration Layer); Domain KPI vocabulary and targets (owned by Domain Pack and Business Instance Layers).
The full reference page, with every canonical specification this layer touches, lives in the documentation repository, which is private while the platform is being built.