Octoryn Platform
Octoryn Observe
Observability for AI actions: what was decided, by which model and policy, and what happened next.
Overview
Octoryn Observe is observability built for AI actions rather than for servers and requests. For each action it captures what was decided, which model and version produced the decision, which policy allowed or blocked it, which tools were invoked and what happened next. It is intended for the operations, risk and engineering teams who are accountable when an AI action reaches into real systems and need to answer "what did it do, and why" without reading raw logs. As a Private preview capability, it turns the governance signals Octoryn already emits into a coherent, queryable view.
How it works
Capture governance signals
As an action moves through identity, policy, tool-access, approval and execution, each step emits a structured signal rather than a free-text log line.
Link into one action record
Signals are correlated by action, so a single view shows the decision, the model and version, the policy outcome and the downstream effect together.
Classify the action
Each action is tagged as read-only, reversible or irreversible, so attention can focus on the effects that matter.
Query and slice
You can filter by model, policy, tool, outcome, approver or classification to see patterns across many actions.
Surface anomalies
Blocked actions, escalations, repeated fail-closed events and unusual tool use are made visible rather than buried.
Key capabilities
Decision-level detail
Each record shows the policy decision and its reason, not only that a call was made.
Model and version attribution
Every action carries the model and version that produced it, so behaviour can be traced to a specific configuration.
Action classification view
Read-only, reversible and irreversible actions are separated, letting review effort concentrate on irreversible effects.
Tool-invocation trail
The tools an action called, with their inputs and outcomes, are recorded as part of the action.
Approval and escalation visibility
Where a human approved or an action escalated, who and when is part of the record.
Cross-action querying
Filters across model, policy, tool and outcome let you spot trends and outliers over time.
What it is not
- It is not a replay — Observe tells you what was decided and what happened, but does not reconstruct a run from its inputs; that is Octoryn Replay.
- It is not a general APM or infrastructure-metrics tool; its unit is the AI action and its governance, not CPU, latency or traces.
- It does not judge whether a decision was correct; it makes the decision and its context legible for humans to assess.
Integration
Observe consumes the governance signals Octoryn already emits along the action path, so where actions are already governed there is little new instrumentation to add. Its records can be exported to your existing log store, SIEM or data warehouse for retention and correlation with non-AI events. Which signals are in scope, retention periods and export destinations are assessed per engagement.
Deployment options
- Managed cloud
- Our cloud account
- Private cloud / VPC
Example use cases
- An operations team investigates why a batch of AI actions were blocked overnight and finds a single policy and model version behind them.
- A risk reviewer filters to all irreversible actions taken last week and confirms each carried the required human approval.
- An engineer compares tool-invocation patterns across two model versions to see whether a change altered which tools actions reach for.
Frequently asked questions
How is Observe different from our existing logging or APM?
Existing tools centre on requests, latency and infrastructure health. Observe centres on the AI action — the decision, the model and version, the policy outcome, the tools and the human approvals — and links them into one record so the "why" is legible, not reconstructed from scattered logs.Can Observe records leave Octoryn for our own systems?
Yes, records are intended to be exportable to your log store, SIEM or warehouse. This lets you retain them under your own policy and correlate AI actions with non-AI events; destinations and retention are set per engagement.Does Observe tell me if the AI made the right call?
No. It makes the decision and its full context visible so your people can assess it. Judging correctness stays with humans; Observe’s job is to remove the guesswork about what happened.
All capabilities
Octoryn Builder
Private previewBuild production applications that emit portable, reviewable source code — not a locked-in low-code runtime.
- Audience:
- Engineering teams and organisations building internal and customer-facing software.
Octoryn Runtime
Private previewA runtime for governed AI-enabled applications, connecting models, policies, tools and operational systems.
- Audience:
- Teams operating AI inside real workflows and systems of record.
Octoryn Gateway
Private previewA controlled AI gateway so model access passes through organisational policy instead of scattered API keys.
- Audience:
- Security, platform and data teams standardising enterprise model access.
