DUBSAR and LangSmith
Alternative · LangSmith
LangSmith and DUBSAR: distinguish traces, durable state and mission decisions
Your team can inspect a model call and compare two versions of an agent. At handover, the client also asks which action was authorised and whether its effect is confirmed. Connect that information to the work while distinguishing observation traces from the state used for execution.
LangSmith and the LangGraph ecosystem already address several of these questions. DUBSAR is developing a governance foundation for agent missions. A fair comparison recognises overlap before looking for complementarity.
In brief
The question and the useful result.
- Question
- Which LangSmith and LangGraph information needs to be linked to a mission’s authorisations and outcomes?
- Useful result
- A reading grid connecting traces, checkpoints, human decisions and confirmations from business systems.
DUBSAR and LangSmith
LangSmith extends beyond tracing
The observability concepts describe runs, traces and threads used to inspect applications. Evaluation tools compare outputs with datasets and criteria, offline or on live traffic.
LangSmith Deployment also documents an orchestration runtime with durable execution. LangGraph persistence describes checkpoints, thread state and workflows involving human input. Reducing the ecosystem to observation without recovery would be inaccurate.
DUBSAR and LangSmith
Identify the evidence each question needs
An observability thread groups traces; execution state supports the running workflow. Specify their roles and relationship in the application concerned. Information carrying the same identifier does not thereby become a business authorisation.
| Information | What to examine |
|---|---|
| Trace and evaluation result | Observed operations, tested version, criteria, covered data and limits. |
| State and checkpoint | Retained data, actual storage and the point from which execution can continue. |
| Human decision | Action, object, reviewed content, authorised decision-maker and changes requiring fresh review. |
| Business effect | Confirmation from the system performing the action, tied to the correct attempt. |
DUBSAR’s intended contribution is a durable relationship between these elements for teams deploying and handing over missions. That relationship remains to be demonstrated in each assembly; it is not an exclusive capability to attribute to DUBSAR.
DUBSAR and LangSmith
Scenario: the version changes while approval is pending
In this illustrative example, a workflow prepares a reply and waits for a responsible person’s agreement. A colleague then changes the recipient and an attachment. The team needs to establish which version was actually reviewed before resuming sending.
A satisfactory evaluation of the initial draft does not automatically cover the new message. A retained checkpoint can recover state according to the implementation; the decision’s scope and the path performing the action still need checking.
If the application receives only an acknowledgement of the sending request, the mission retains the outcome as uncertain. Recovery should inspect available evidence before another attempt. This scenario describes neither a universal LangGraph limitation nor a qualified DUBSAR connector: it sets out checks for the selected workflow.
DUBSAR and LangSmith
Prepare operations and handover
Name the source of execution state and the source of decisions. Retain accessible links between mission, version, checkpoint, trace and business attempt, with suitable access and retention. Establishing those relationships does not require copying every prompt.
Examine the persistence actually configured and its behaviour after a restart. Test refusal, changes after approval and uncertain external outcomes alongside normal success. Agent-evaluation criteria and action-authorisation criteria should remain identifiable.
Retain mechanisms that already meet the need in LangSmith, LangGraph or the application. Adding DUBSAR should not be presented as necessary in principle. No integration between these products is announced here.
DUBSAR and LangSmith
Frequently asked questions
Is LangSmith only an observability tool?
No. The reviewed sources also cover evaluation and LangSmith Deployment. LangGraph provides persistence and workflows involving human input. Compare the scopes actually used.
Does a checkpoint prove that an external action succeeded?
It retains state according to the configured mechanism. Establish the external result through an interpretable response from the relevant system, linked to the correct operation. Local state does not replace that check.
Must LangGraph be replaced to consider DUBSAR?
No need for replacement has been established. Possible complementarity would require defined responsibilities, interfaces and expected evidence, followed by qualification of the relevant workflow.
DUBSAR and LangSmith
Where DUBSAR fits
DUBSAR is developing a governance foundation for integrators and teams deploying agents for clients: durable mission state, explicit authorisations, human decisions, evidence and continuation of work. It aims to connect what was requested, what was permitted and what is established about the outcome.
Selected components and workflows have been tested; full B2B integration and validation are still underway. Pilots are not currently open.
DUBSAR and LangSmith
Sources and scope of review
Official sources reviewed on 20 October 2026. Scenarios and records are preparation methods that can be used without DUBSAR. They establish neither an available integration nor compliance.
- LangSmith — Observability concepts — Observability runs, traces and threads.
- LangSmith — Evaluation — Offline and online evaluations, datasets and criteria.
- LangSmith — Deployment — Orchestration runtime, durable execution and deployment options.
- LangGraph — Persistence — Checkpoints, thread state, recovery and storage choices.
DUBSAR and LangSmith