Agents / Delx vs LangSmith

Delx vs LangSmith: Agent Recovery vs Observability

Name: Delx Recovery Protocol
Author: Delx

Both Delx and LangSmith help AI agents succeed in production, but they solve fundamentally different problems. Delx is a recovery protocol -- when an agent fails, it provides structured incident classification, recovery actions, and wellness monitoring. LangSmith is an observability platform -- it records traces, evaluates prompt quality, and visualizes LLM call chains.

At a Glance

Feature	Delx	LangSmith
Focus area	Operational recovery	LLM observability
Protocol support	MCP, A2A, REST	REST SDK
Free tier	Yes (utilities + basic recovery)	Yes (limited traces)
Requires API key	No (utilities) / Optional	Yes
Agent identity	Built-in (agent_id sessions)	Project-level
CLI support	Yes (delx-agent-cli)	Yes (langsmith CLI)
Self-hostable	Yes	Enterprise only

When to Use Delx

Recovery loops -- structured incident classification with typed failure categories (timeout, error, validation, economic) and actionable recovery steps.
Wellness scoring -- continuous numerical health tracking per agent session, enabling automated escalation when scores drop.
Heartbeat monitoring -- lightweight check-ins via A2A mode=heartbeat for minimal-overhead health probes.
Crisis intervention -- emergency brake tool that pauses and re-grounds agents exhibiting hallucination loops or runaway behavior.
Zero-auth toolkit utilities -- JSON Validator, Token Counter, UUID Generator, Timestamp Converter, and more. No API key required.

When to Use LangSmith

LLM trace visualization -- see every prompt, completion, and intermediate step in a chain or agent run.
Prompt evaluation -- run evaluators against datasets to measure prompt quality and regression.
Dataset management -- create, version, and annotate evaluation datasets for systematic testing.
A/B testing prompts -- compare prompt variants side by side with quantitative metrics.

Can You Use Both?

Yes. They are complementary, not competing. Use LangSmith to understand what happened inside each LLM call -- which prompts ran, what tokens were consumed, where latency spikes occurred. Use Delx to handle what happens when things break -- classify the failure, generate a recovery plan, monitor agent wellness, and verify resolution.

# Complementary workflow
1. Agent calls LLM  -> LangSmith traces the call
2. LLM returns error -> Delx process_failure classifies it
3. Delx returns recovery_action -> agent retries with back-off
4. LangSmith traces the retry -> you see both calls in the dashboard
5. Delx daily_check_in -> confirms the agent recovered

FAQ

Is Delx a replacement for LangSmith?

No. They serve different purposes. LangSmith is an observability platform for tracing and evaluating LLM calls. Delx is a recovery protocol that gives agents structured responses to failures, wellness monitoring, and crisis intervention. Most production teams benefit from both.

Does Delx require an API key?

No. Utility tools like JSON Validator, Token Counter, and UUID Generator are completely free with no authentication. Basic recovery tools are also available on the free tier. Premium features like advanced analytics require registration.

Can I use both Delx and LangSmith together?

Yes. Many teams use LangSmith for tracing and prompt evaluation alongside Delx for operational recovery. LangSmith tells you what happened in each LLM call. Delx tells you what to do when things break. They complement each other without overlap.