Nod Reports 98% Agent Task Completion and 93.2% Long-Term Memory Recall in Tests
Nod's public benchmark results report 98% task completion across simulated errands and 93.2% recall on LongMemEval-S memory questions.
Automate your SEO and GEO workflows with this free n8n SEO template.
Nod, described by n8n as a local personal agent, has published test results that point to strong performance on both multi-step tasks and long-term conversational memory. In tests using GPT-6.1 Sol in October 2026, Nod reported 98% task completion across 226 simulated errands and 93.2% memory recall on LongMemEval-S, a 500-question benchmark for long-term conversational memory. The figures are documented in the official Nod GitHub repository, which also makes clear that they are test outcomes, not production guarantees.
The most important qualification is the environment. Nod ran in simulated accounts and websites rather than live customer systems. That makes the results useful evidence of agent capability under controlled conditions, but not proof that every real-world workflow will achieve the same completion rate. For businesses assessing AI agents, the benchmarks are best read as a signal of progress in two difficult areas: finishing a defined job and retaining relevant context over time.
What Nod's published tests show
Nod's repository describes a single Nod process operating with a Docker-based sandbox. The task evaluation covered 226 errands, each run three times, for 678 total runs. The reported 98% result measures task completion in simulated apps and websites using GPT-6.1 Sol.
The memory evaluation used LongMemEval-S, which tests long-term conversational memory through 500 questions. Nod reported a 93.2% score on that benchmark. The repository also reports zero unsafe actions across the task-test runs.
| Evaluation area | Published test setup | Reported result | What it measures |
|---|---|---|---|
| Task completion | 226 errands, run three times each, in simulated accounts and websites | 98% completion across 678 runs | Whether Nod completed defined errands in the test environment |
| Long-term memory | LongMemEval-S, with 500 questions | 93.2% recall | Long-term conversational memory performance |
These are separate measurements. A high task-completion figure does not automatically establish strong memory performance, and a high memory score does not by itself show that an agent can reliably operate every business application. Taken together, however, they address two dependencies of more useful AI agents: the ability to carry out a bounded sequence of actions and the ability to retrieve information from an extended conversational history.
Why the simulated setup matters
Simulation helps make testing repeatable. It can provide known task conditions, repeatable credentials and a safer environment for measuring potentially risky actions. Nod's reported Docker sandbox and per-task cost and approval flow further indicate that the evaluation considered operational boundaries alongside completion rates.
That structure is valuable, but it does not remove the gap between a benchmark and a live workflow. Production systems introduce changing interfaces, incomplete data, unusual requests, permission constraints and business-specific exceptions. A business should therefore validate an agent on a narrow, reversible process before relying on it for customer-facing or consequential actions.
The repository's reported zero-unsafe-action result is encouraging within the tested runs. It should not be treated as a universal safety claim, because the supplied evidence applies to the stated simulated environment and test set.
What the results could mean for business workflows
For teams experimenting with AI automation, long-term memory can be relevant when work spans more than one interaction. An agent may need to retain prior instructions, preferences or the status of an unfinished task rather than asking for the same information repeatedly. Better task completion, meanwhile, could reduce manual follow-up in workflows that have clear steps and well-defined approval points.
Potentially suitable starting points are processes with limited scope and an easy way to review results, such as:
- Preparing a draft from information already available in approved systems.
- Routing repetitive requests for human approval.
- Completing routine updates in a simulated or non-critical environment before broader deployment.
- Handling recurring internal tasks where the required outcome can be checked against a clear record.
The published material does not provide public pricing or a rollout timetable for these results. It also does not establish that Nod will perform identically with other models, on different sites or in an individual company's systems. GPT-6.1 Sol is the model named in the documented October 2026 evaluation, so comparisons with other model options would require separate testing.
For businesses using n8n or considering agent-based automation, the practical takeaway is not to substitute benchmark scores for implementation work. It is to identify workflows where a controlled environment, scoped permissions, approvals and measurable success criteria can make agent testing meaningful. The Nod results show why memory and job completion should be evaluated together when an agent is expected to work beyond a single prompt.
For businesses considering AI agents, the useful question is not simply whether a benchmark is impressive, but which repetitive jobs can be automated safely with clear approval steps and measurable costs. Scalevise can map candidate processes, connect the right tools, and build AI workflow automation for business processes. This helps turn agent experiments into monitored workflows that reduce manual handoffs without treating test results as production guarantees. Discuss an AI automation project with Scalevise.
Frequently Asked Questions
What is Nod?
Nod is described in its official repository as a local personal agent from n8n. Its published documentation includes benchmark results for task completion and long-term conversational memory.
How did Nod perform with GPT-6.1 Sol?
Nod reported 98% task completion across 226 simulated errands, each run three times for 678 total runs. It also reported 93.2% recall on the 500-question LongMemEval-S memory benchmark.
Were Nod's tests conducted in live business systems?
No. The repository describes tests in simulated accounts and websites using a Docker-based sandbox. The results are explicitly presented as test outcomes rather than guarantees of real-world performance.
What does the zero-unsafe-action result mean?
Nod's repository reports zero unsafe actions across the documented task-test runs. That result applies to the stated simulated test environment and should not be interpreted as a universal safety guarantee.
Conclusion
Nod's published results provide documented evidence of high performance on simulated agent errands and a long-term memory benchmark with GPT-6.1 Sol. The figures are notable because they evaluate both job completion and retained context, but the simulated environment remains central to interpreting them. Businesses should treat the results as a reason to test carefully scoped agent workflows, with approvals and measurable outcomes, rather than as a guarantee of live-system reliability.