Avesha
Also known as: Obliq
Obliq, Avesha's autonomous AI SRE for Kubernetes and agentic workloads: agents watch CI and telemetry, synthesize root cause and execute rollbacks, replica bursts or config patches under risk scores and policy as code.
Obliq is Avesha's autonomous AI SRE for Kubernetes and agentic workloads, built on the Avesha platform and its KubeSlice connectivity layer. A Code-Drift Analyzer watches CI, pull requests and AI generated commits to flag untested diffs before rollout, a Root-Cause Synthesizer correlates logs, traces and metrics, including model telemetry such as token latency and prompt drift, into a causal narrative, and an Auto-Remediator executes safe rollbacks, replica bursts or configuration hot patches, with declarative playbooks triggering validated fixes.
Governance is policy based: Obliq calculates risk scores, enforces custom SLAs and policies as code, applies time bound access controls and integrates RBAC with existing identity providers, with audit hooks for compliance. Telemetry is stitched with memory of prompt, tool and outcome, and incidents feed reinforcement learning loops. Pricing is not published and runs through a sales conversation.
Vendor details
Canonical URL
https://avesha.io
Category
SRE / DevOps agent
Funding status
Venture backed
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Integrates telemetry through MCP services for AWS, Prometheus, Loki, Neo4j, and CloudWatch, plus Slack, DataDog, Kubernetes events, and OpenTelemetry; installs via Helm charts into the customer's cluster.
In practice
An AI generated commit slips past review. Obliq's Code-Drift Analyzer watches CI and pull requests and flags untested diffs before rollout.
A deploy degrades latency across services. Obliq synthesizes the root cause from logs, traces and metrics and can roll back or burst replicas under your policies.
Your AI workloads misbehave in ways dashboards miss. Obliq tracks token latency, context window overflows and prompt drift alongside infrastructure metrics.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool Calling | Full |
|---|---|
|
Obliq's Auto-Remediator executes rollbacks, replica bursts and configuration hot patches on the customer's Kubernetes, and its agents read metrics, traces, logs and CI and pull request activity. RBAC integrates with existing identity providers. Sourceavesha.io/obliqread 2026-09-29 |
|
| Workflow Orchestration | Full |
|
Three agents hand work along a chain, a Code-Drift Analyzer, a Root-Cause Synthesizer and an Auto-Remediator, and declarative playbooks trigger validated rollbacks or auto-patches under custom SLAs and policies. Sourceavesha.io/obliqread 2026-09-29 |
|
| Knowledge Grounding & RAG | Partial |
|
Root cause work correlates live metrics, traces, logs, slice health and model telemetry into a causal narrative per incident. There is no maintained index over runbooks or knowledge bases. Sourceavesha.io/obliqread 2026-09-29 |
|
| Human Oversight & Guardrails | Partial |
|
Customer controlled constraints govern what the agent may do. Obliq calculates risk scores, enforces custom SLAs and policies through policy as code, and applies time bound access controls. There is no step where a person approves an action before it runs. Sourceavesha.io/obliqread 2026-09-29 |
|
| Security, Identity & Governance | Partial |
|
Access runs on policy as code and RBAC integration with existing identity providers, with time bound access controls. There is no published certification or trust center. Sourceavesha.io/obliqread 2026-09-29 |
|
| Observability & Auditability | Partial |
|
Each incident gets a causal narrative, and Obliq offers audit hooks for internal compliance, but Avesha does not describe a per run record of the steps the agents took or the actions the Auto-Remediator executed. Sourceavesha.io/obliqread 2026-09-29 |
|
| Memory & State Persistence | Partial |
|
Telemetry is stitched with memory of prompt, tool and outcome, and Obliq learns from prior decisions to improve outcomes, but Avesha gives that memory no stated scope, lifetime or customer control. Sourceavesha.io/obliqread 2026-09-29 |
|
| Deployment & Data Residency | Full |
|
Obliq needs a Kubernetes cluster, Helm 3.8 or later and cluster admin access, and it deploys as a single Helm chart into the customer's own cluster, with integration pages for AWS, Google Cloud and Oracle Kubernetes Engine. SourceAvesha Obliq 1.5.0 prerequisites and repo.obliq.avesha.io; docs.avesha.io/documentation/enterprise-obliq/1.5.0/prerequisitesread 2026-09-30 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
Named prebuilt agents each have their own job. The Code-Drift Analyzer flags untested diffs from CI and pull requests on its own, beside the Root-Cause Synthesizer and the Auto-Remediator, with declarative playbooks. Sourceavesha.io/obliqread 2026-09-29 |
|
| Triggers & Channel Coverage | Full |
|
Obliq watches CI, pull requests and GenAI commits to flag untested diffs before rollout, and acts on telemetry anomalies and policy violations in real time, detecting and self healing before humans are paged. Sourceavesha.io/obliqread 2026-09-29 |
|
| Model Flexibility & Routing | Full |
|
The Obliq 1.5.0 integrations let the customer configure OpenAI with its own API key and choose a model from a dropdown, configure Google Gemini the same way, or point the agent at any endpoint following the OpenAI chat completions format with its own key and a custom model ID. So the customer chooses across model makers and can bring its own model. Sourcedocs.avesha.io/documentation/enterprise-obliq/1.5.0/integrations/openai-compatibleread 2026-09-30 |
|
| APIs / SDKs / MCP Extensibility | Partial |
|
Obliq is called MCP-Ready, but no endpoint, auth scheme or tool list is published, and there is no API or SDK for Obliq. The MCP services connect outward to the customer's AWS, Prometheus, Loki and CloudWatch and do not give outside callers a way in. Sourceavesha.io/obliqread 2026-09-29 |
|
| Testing, Debugging & Optimization | Partial |
|
Avesha describes a learning loop in one sentence, in which incidents feed reinforcement learning loops and the system upgrades itself. Flagging untested diffs tests the customer's code, not the agent, and there is no evaluation harness. Sourceavesha.io/obliqread 2026-09-29 |
|
| Browser / Computer-use | Not documented |
|
Obliq acts on Kubernetes and cloud through APIs and playbooks. There is no browser, desktop or computer control. Sourceavesha.io/obliqread 2026-09-29 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Enterprise pricing; not publicly listed
Enterprise engagements scoped to environment size and workloads; self hosted install with provisioned credentials.
Cost watchouts
The agent requires the customer's own OpenAI API key, so model usage is billed separately; cost scales with monitored infrastructure and GPU workloads.
Variable cost rationale
Cost scales with cluster and environment size, monitored workloads, and GPU usage; the platform also requires the customer's own model API keys, adding usage cost. No public rate is available.
Sales call required
Yes, required for paid access
Free / trial
Credentials are provisioned through Avesha; no public self serve tier.
Lowest paid plan
Not publicly listed.
Key ambiguities
No public pricing; the self hosted model means model API usage and infrastructure cost sit on top of any license.