Conversational IT support
Interpret support requests across access, connectivity, software, and hardware workflows.
IT support and telemetry intelligence
Podotl
Understand the ticket.
Connect the telemetry.
Bring conversational IT support and machine signals into the same diagnostic workflow. Podotl uses a gated state-space model to support investigation and remediation planning inside your environment.
Two inputs. Shared context.
Connect the service desk to infrastructure.
Machine telemetry
Work with logs and numerical metrics from cloud, virtualisation, containers, networks, and operating systems.
Constrained remediation
Evaluate proposed actions against declared invariants, including capacity floors and allowed changes, before execution.
Model architecture
A recurrent model for operational signals.
Podotl uses a gated linear state-space model (GL-SSM). A compact recurrent state carries context forward while input-dependent gates control what is retained and updated.
Fused signals
Project numerical telemetry into model representations alongside log and support-ticket tokens.
Diagnostic and planning heads
Generate diagnostic outputs and remediation proposals from the shared operational representation.
Neuro-symbolic checks
Use Z3/Presburger verification to check encoded preconditions and postconditions for supported action contracts.
State update and verification boundary
h_t = alpha_t ⊙ h_(t−1) + beta_t ⊙ (W_in x_t)
Input-dependent decay and selectivity update a fixed-size recurrent state. This avoids a growing attention KV cache for the recurrent model state.
Formal checks apply to the encoded contract and its assumptions. Operational permissions, action adapters, and post-action health verification remain part of the surrounding system.
Engineering evaluation
Different tests answer different questions.
Internal evaluations on NVIDIA GeForce GTX 1650 hardware. Batch evaluation, HTTP generation, and scenario matching are separate workloads.
| Evaluation | Workload | Reported result |
|---|---|---|
| Tensor batch evaluation | 10,000 incidents; batch size 32 | 189.7 incidents/s; 52.72 seconds |
| HTTP generation | 100 requests; 8 concurrent workers | 13.91 s median; 22.82 s p95; 72% keyword/remediation match |
| Critical scenario suite | 25 scenarios across 23 domains | 25/25 passed report criteria; 77.3% average keyword match; 2.36 s average inference |
Sources: DevHeal 10,000-Incident Parallel Benchmark and Real-Life Scenario Test reports. Batch throughput is not interactive request latency. Scenario pass counts are not a claim of universal diagnostic accuracy.
Your workflow. Your first autonomous outcome.
See autonomy in action.
Start with a free pilot.
Curious how autonomous behaviour could work for your business? Bring a real challenge. We’ll help you define a focused pilot, explore the right product, and agree what success looks like in your environment.
Request your free pilot →
No guesswork. Start with a conversation.