Find the problem when it isn't the LLM.
When an agent run goes wrong, see whether the cause is the LLM or a service underneath it. DeepCube shows the services around the same trace.
Is DeepCube a LangSmith alternative?
For seeing agent runs in the context of your whole application, yes. For evaluation, the prompt playground, agent deployment, alerting and paging, no. Keep LangSmith for those and run DeepCube next to it. Visualize takes 200 traces and logs per second for the whole grid, so check that against your volume. DeepCube is a Windows app today, with an alpha on macOS.
Intro
LangSmith is a strong LLM tracing and evaluation platform. It is framework agnostic and now ingests OpenTelemetry from any framework, with deep tooling for active LLM development: prompt playground, annotation queues, dataset management, and detailed trace visualization. We respect that. This page is an honest look at where DeepCube takes a different approach, where LangSmith excels, and how you can evaluate both. The difference is not which frameworks each tool sees; it is whether your LLM traces are correlated with the full application and service topology underneath them.
DeepCube is the product; spatial observability is the category. We are not an APM company, we are a spatial observability company.
Any LLM framework. Full application context. One view.
Deep Agent Tracing, Then What Sits Underneath
What happens when the problem is below the LLM layer?
- LangSmith traces agents, and production AI applications rarely stay within one framework. You might use LangChain for orchestration, a custom retrieval pipeline, and a fine-tuned model served via vLLM. LangSmith ingests OpenTelemetry from any framework via its OTLP endpoint and shows those spans as traces. (source: docs.langchain.com/langsmith/trace-with-opentelemetry, verified 2026-11-06)
- Built for agent work: LangSmith's docs define its data as runs, traces, threads and trajectories. Deployment metric charts cover agents you host on LangSmith Deployment. (source: docs.langchain.com/langsmith/observability-concepts and docs.langchain.com/langsmith/control-plane#monitoring, verified 2026-11-06)
- Open SDK, proprietary platform: LangSmith's client SDK is open source (MIT) and it accepts end-to-end OpenTelemetry ingestion from any OTel-instrumented app, so instrumentation is portable. The platform itself is proprietary software. The differentiator is not framework coverage; it is correlating LLM traces with full application and service topology. (source: github.com/langchain-ai/langsmith-sdk and docs.langchain.com/langsmith/faq, verified 2026-11-06)
Framework-agnostic. Application-level correlation. No lock-in.
Framework-Agnostic, Application-Level Observability
Works with any LLM framework. Correlates with everything else.
- Built on OpenTelemetry, not on any single LLM framework. DeepCube works with LangChain, LlamaIndex, Semantic Kernel, custom pipelines, or any combination.
- Your LLM traces live alongside your application metrics and service topology in a single 3D spatial environment.
- When your LangChain agent slows down, DeepCube shows you whether the issue is in the chain logic, the model provider, the retrieval layer, or the application underneath.
No framework lock-in. Application-level correlation.
Architecture: How We Differ
Open standard ingestion vs a proprietary platform.
LangSmith is framework agnostic and ingests OpenTelemetry from any framework. The prompt playground, annotation queues, and dataset management are built for teams doing active LLM development. Its documented data model is agent work: runs, traces, threads and trajectories.
DeepCube works with any LLM framework via OpenTelemetry too. The difference is correlation: it provides deep tracing for LangChain, LlamaIndex, Semantic Kernel, or custom pipelines, and ties that LLM telemetry to your full application health and 3D service topology.
| Aspect | DeepCube | LangSmith |
|---|---|---|
| Scope | Spatial observability: LLM observability correlated with full application context | Agent engineering platform: LLM tracing and evaluation, plus agent deployment, sandboxes, an LLM gateway, no-code agents, and a fix-proposing agent. Built for agents: "every step your AI agent takes". (source: langchain.com/pricing and docs.langchain.com/langsmith/observability-concepts, verified 2026-11-06) |
| Framework Support | Any OTel-compatible framework | Framework agnostic; ingests OpenTelemetry from any framework via its OTLP endpoint (source: docs.langchain.com/langsmith/trace-with-opentelemetry, verified 2026-11-06) |
| Instrumentation | Standard OpenTelemetry SDKs | LangSmith SDK (open source, MIT); also accepts end-to-end OTel ingestion (source: github.com/langchain-ai/langsmith-sdk, verified 2026-11-06) |
| Visualization | 3D spatial topology | Trace trees, prebuilt and custom dashboards with configurable charts (source: docs.langchain.com/langsmith/dashboards, verified 2026-11-06) |
| Service Topology | Auto-discovered 3D service map | Data model covers agent work: runs, traces, threads, and trajectories. OTel fanout also sends spans to Datadog or Grafana (source: docs.langchain.com/langsmith/observability-concepts and docs.langchain.com/langsmith/trace-with-opentelemetry, verified 2026-11-06) |
| Application Metrics | Application metrics via OTel correlated with traces | OTLP endpoint takes traces. Deployment charts cover agents hosted on LangSmith Deployment (CPU, memory, replicas, Postgres, queue depth). (source: docs.langchain.com/langsmith/trace-with-opentelemetry and docs.langchain.com/langsmith/control-plane#monitoring, verified 2026-11-06) |
| Prompt Management | In DeepCube, via Tessa in your codebase workspace | Prompt playground, versioning, hub (source: langchain.com/pricing and docs.langchain.com/langsmith/manage-prompts, verified 2026-11-06) |
| Annotation / Evaluation | Not a DeepCube feature | Annotation queues, dataset management, online evals (source: langchain.com/pricing, verified 2026-11-06) |
| Vendor Lock-in | Low. Change one endpoint and its API key header to leave. | Low: LangSmith ingests OpenTelemetry from any source, and your Collector can send the same spans to Datadog or Grafana. Bulk export is Enterprise-only for new customers (source: docs.langchain.com/langsmith/trace-with-opentelemetry and docs.langchain.com/langsmith/data-export, verified 2026-11-06). LangSmith itself is proprietary software; self-hosted and hybrid deployment are Enterprise-only (source: docs.langchain.com/langsmith/faq and langchain.com/pricing, verified 2026-11-06). |
AI that works in your code is table stakes. What it investigates from is not.
AI: What Each Assistant Investigates From
Code access is table stakes. Context is not.
Both products now ship AI that works in your code. LangSmith Engine reads a connected repository and opens pull requests on Plus and Enterprise; its fix validation and Red Teaming features are still in private beta. In DeepCube, Tessa works in your connected codebase workspace, on every DeepCube plan. You own the outcome.
The difference is what each one investigates from. Engine diagnoses against traces and connected source code. In DeepCube, Tessa investigates from where you stand in the 3D model of your application: from the node or trace behind a degraded LangChain result, Tessa helps you read the traces around it and proposes a fix.
| Capability | Tessa (DeepCube) | LangSmith |
|---|---|---|
| AI Diagnosis | In DeepCube, Tessa helps you read what the traces show, with spatial context | Engine (autonomous issue detection, root-cause diagnosis, and proposed fixes; Plus and above; its fix validation and Red Teaming features are in private beta, LangSmith Cloud only) (source: docs.langchain.com/langsmith/engine and docs.langchain.com/langsmith/engine-fix-validation, verified 2026-11-06); Insights (automated topic clustering with optional guided categories, scheduled reports) (source: docs.langchain.com/langsmith/insights, verified 2026-11-06); Remote MCP exposes prompts / traces / datasets / experiments to external AI agents (source: docs.langchain.com/langsmith/langsmith-remote-mcp, verified 2026-11-06) |
| Codebase Access | In DeepCube, your local checkout: read, search, propose edits | LangSmith Engine connects GitHub repositories via a GitHub App, or GitLab projects on Cloud via a token, and reads the source to locate the code path behind a failing trace. Plus and Enterprise only (plans, verified 2026-11-10). (source: docs.langchain.com/langsmith/engine-repositories, verified 2026-11-06) |
| What the AI can reach in your source | In DeepCube, Tessa reads your code where it already lives, on your machine. No vendor GitHub App, and no copy of your repo stored in our cloud. What Tessa reads is relayed to the model for your question and not kept on our servers. By default the model sees your workspace's top-level file names and any agent instructions file the repo ships. Other file contents go only when she opens them. (Tessa docs, data security) | LangSmith Engine connects through a GitHub App or a GitLab token. With source access, it "locates the code path behind a failing trace", in the repositories you select (source: docs.langchain.com/langsmith/engine, verified 2026-11-06) |
| Code Fixes | Available on every plan (on Start, within Tessa's free allowance of 1 request per 6 hours). In DeepCube, Tessa helps you follow the issue to the source file in your connected workspace and proposes the change. You own the outcome. | LangSmith Engine opens a pull or merge request with proposed code changes when you click Open PR. Plus and Enterprise only. Engine's fix validation and Red Teaming are in private beta on LangSmith Cloud, by waitlist. Engine is metered in LSUs (1 LSU = $1; an Engine run can consume about 7 to 45 LSUs), an additional charge on top of the Plus seat fee. (source: docs.langchain.com/langsmith/engine-issues, docs.langchain.com/langsmith/engine-fix-validation, langchain.com/pricing, verified 2026-11-06) LangSmith Chat (formerly Polly) refines prompts and generates evaluator code. (source: docs.langchain.com/langsmith/chat, verified 2026-11-06) |
| Root Cause Analysis | In DeepCube, Tessa works across LLM, app and dependencies | Engine diagnoses "against your traces and connected source code." (source: docs.langchain.com/langsmith/engine, verified 2026-11-06) |
| Accountability Model | In DeepCube, Tessa works through your traces, logs and code on her own, then brings you the answer. You own the outcome | On Plus and above, Engine detects, diagnoses and proposes fixes on a schedule and reopens recurring issues. A person clicks Open PR, and Engine never merges. (source: docs.langchain.com/langsmith/engine and docs.langchain.com/langsmith/engine-issues, verified 2026-11-06) |
How each one bills.
What you would pay, and for what
How each one bills.
- What each covers: DeepCube includes LLM tracing, application traces and logs, and Tessa on one primary meter (nodes). LangSmith now spans Observability, Evaluation, Deployment, Fleet, Engine, Sandboxes, and LLM Gateway, billed in seats and LSUs (1 LSU = $1). (source: langchain.com/pricing, verified 2026-11-06)
- Priced by node: DeepCube works with any LLM framework via OpenTelemetry and prices by node, not by seat or trace.
- Predictable pricing: Node count sets the bill, with a volume discount on Visualize. The platform is priced per node, not per seat, and there are no per-trace charges.
- AI included: Tessa is part of every DeepCube plan at no extra charge. AI is included in the per-node price: you need no AI seat, and there is no per-query AI charge. Optional Tessa Personal seats (Tessa E5, E25 and E50 on the pricing page) add capacity. On paid plans the bundled Tessa allowance is a per-node budget; when it is spent, Tessa keeps working on a lighter model rather than stopping. Every plan can run, in DeepCube, Code Fix and root-cause analysis (on Start, within Tessa's free allowance of 1 request per 6 hours); higher plans put a more capable model behind them, and Fuse, available now through sales, adds voice and agents. LangSmith meters Engine in LSUs on top of the Plus seat fee. (source: langchain.com/pricing, verified 2026-11-06)
| Capability | DeepCube | LangSmith |
|---|---|---|
| Bills on | Per application instance (a node) | Per seat, then pay as you go. Usage is metered in LangChain Standard Units (LSUs) (source: langchain.com/pricing, verified 2026-11-06) |
| List rate | Visualize: $20 per node, with a volume discount | Developer free / Plus $39/seat/mo / Enterprise custom. 1 LSU = $1 (source: langchain.com/pricing, verified 2026-11-06) |
| In the price | Logs and Tessa | Developer: 1 seat, 5k base traces a month. Plus: 10k base traces a month and one free small serverless deployment (source: langchain.com/pricing, verified 2026-11-06) |
| Not in the price | Infrastructure monitoring, alerting and paging. Optional: Tessa Personal seats (Tessa E5, E25 and E50), support hours. A $99.99 fee applies when you change plan. | Traces above the included amount are pay as you go. On Plus, LSUs are an additional charge on top of the seat fee (source: langchain.com/pricing, verified 2026-11-06) |
| Plan limits | Visualize: 200 traces and logs per second for the whole grid, 30-day retention, shared infrastructure. Start: 50 per second, 7-day retention. Telemetry above the limit can be dropped, and your collector will not show an error. DeepCube is a Windows app today, with an alpha on macOS. | Developer: 1 seat. Plus: unlimited seats. Base traces keep 14 days, extended traces 180 days (source: langchain.com/pricing, verified 2026-11-06) |
| What it is for | Runs next to your current tool for tracing failures. No self-serve plan is meant to stand in for it today. Analyze and Fuse are available now through sales, with node minimums of 25 and 50. | "Debug, monitor, and improve your AI applications with tracing and evals" (source: langchain.com/pricing, verified 2026-11-06) |
| Your number | Count your application instances, then see deepcube.ai/pricing | langchain.com/pricing |
| Distributed Tracing | Included | Available across services via context propagation headers (source: docs.langchain.com/langsmith/distributed-tracing, verified 2026-11-06) |
| Application Traces | Included | Agent Server APM tracing goes to Datadog or an OTLP target (OTel path in Alpha) (source: docs.langchain.com/langsmith/env-var-cloud, verified 2026-11-06) |
| AI Assistant | Included (Tessa) | LangSmith Chat (formerly Polly), Insights, and Engine. Insights and Engine require Plus or above; Chat requires you to supply your own model API key as a workspace secret (source: docs.langchain.com/langsmith/chat, verified 2026-11-06) |
| 3D Spatial Topology | Included | Trace trees and dashboard charts (source: docs.langchain.com/langsmith/dashboards, verified 2026-11-06) |
DeepCube pricing from deepcube.ai/pricing. LangSmith pricing from langchain.com/pricing. Verify current pricing before purchase. All prices USD.
It runs next to what you have
Already Using LangSmith? Add Application Context.
It adds to what you have.
- Keep LangSmith for LLM development: Prompt playground, annotation queues, dataset management. Add DeepCube for application traces and logs, and cross-signal correlation.
- Run it alongside: Use DeepCube next to LangSmith. Analyze and Fuse are available now; contact sales.
- Full-stack correlation: As your stack grows, DeepCube ties LLM telemetry to your service topology and application metrics.
- Easy exit: If DeepCube is not right for you, change one endpoint and its API key header. Your instrumentation stays exactly the same.
OTel Collector Config
receivers:
otlp:
protocols:
grpc:
http:
exporters:
otlp/deepcube:
endpoint: "https://otlp.deepcube.ai"
headers:
API-Key: "YOUR-API-KEY"
otlphttp/langsmith:
endpoint: "https://api.smith.langchain.com/otel"
headers:
x-api-key: "YOUR-LANGSMITH-API-KEY"
service:
pipelines:
traces:
receivers: [otlp]
exporters: [otlp/deepcube, otlphttp/langsmith]
metrics:
receivers: [otlp]
exporters: [otlp/deepcube]
logs:
receivers: [otlp]
exporters: [otlp/deepcube]
Standard OTel Collector config. Traces go to both destinations. Metrics and logs go to DeepCube.
Ready to Correlate LLM Traces With Your Whole Application?
Start free with DeepCube. Your OTel instrumentation stays as it is: send it through a Collector or any OTLP/gRPC exporter.
Compare DeepCube against other tools | LLM observability comparison