Compare
While. vs LangChain.
LangSmith watches the agent and edits its prompts and code. While. learns from production traces and continuously improves the model with mid-training and post-training SFT and RL.
Why you need While.
Monitoring tells you the agent failed. It does not make the agent stop failing.
The fix is a change to the model. That takes three things: training data from the failures, a training run, and a held-out evaluation with an interval on the difference.
While. does all three. The other tools on this page stop at the red check.
What LangChain is
LangChain makes the open-source frameworks LangChain, LangGraph and Deep Agents, all MIT, and sells LangSmith, a platform for agents in production. LangSmith traces every run, from its own SDKs or any OpenTelemetry source, builds dashboards and alerts on them, and scores runs with evaluators: code, an LLM judge, human review, or a pairwise comparison of two versions. Datasets come from hand curation, production traces or synthetic generation, and experiments run the agent against them. Deployment hosts the agent server. Engine reads the traces, finds recurring failures, writes prompt and code changes, and opens a GitHub pull request.
There is also a sandbox for agent-written code, an LLM gateway, and Fleet, a no-code agent builder. Plans start free at 5,000 traces a month; self-hosting is an Enterprise add-on. It raised $125 million in a Series B led by IVP in October 2025 at a $1.25 billion valuation. Its site names Expedia, Autodesk, Workday, Nvidia, ServiceNow, Coinbase, Monday.com and Lyft among its customers.
What While. is
While. is a mid-training and post-training platform for language models, built for the team that owns a company's models. The open-source SDK builds and optimizes the dataset: it learns from production traces, writes situations from the agent's tools, plays the agent against them, and scores every run with a judge you write as a program. A hosted trainer runs mid-training and post-training on an open model with the latest SFT and RL algorithms, the model is served on your own endpoint, and a held-out set with a confidence interval says whether the change is real. Curb a behavior, build character, make tool use more efficient, harden the model against adversarial attacks.
The SDK is open source under Apache 2.0 and runs offline with no key. The platform is hosted. We do not publish customer names yet. Public results: a telecom support agent, trained to hand customers off at the right time, went from 5% to 30% on the held-out split of τ²-bench, and a refund agent, trained to refund only within policy, went from 18% to 85% on a held-out set.
Side by side
| LangChain | While. | |
|---|---|---|
| What it does | Traces the agent, scores it on datasets, hosts the agent server, and has Engine write prompt and code fixes from production failures. | Tests the agent, turns the failures into training data, trains a model, serves it, and measures the change on a held-out set with a confidence interval. |
| Open source | The frameworks LangChain, LangGraph and Deep Agents, MIT. LangSmith is closed; self-hosting needs an Enterprise plan. | The whileai SDK under Apache 2.0. The platform is closed. |
| Can you use it today | Yes. A free Developer plan with 5,000 traces a month. Engine and Fleet have a get-started path and a demo request. | Yes. pip install whileai. It runs offline with no key. |
| Where the test comes from | A dataset you curate, cut from production traces, or generate. Online evaluators score live traffic too. | Your production traffic, or situations the SDK writes from your agent's tools when you have no traffic yet. |
| What the agent runs against | Your real agent, called with the dataset inputs. A sandbox is available for code the agent writes. | Fake tool backends built from your tool descriptions, or your real tools inside one function. |
| Judges | Code, an LLM judge, human review, or pairwise. Written by you or picked from built-ins. | A judge is a program over the tool calls. Before you trust its scores, the SDK checks it against hand labels and probes it for length bias and other easy tricks. |
| The number you get | Scores per evaluator per experiment, compared across versions in a table. No interval on the pages we read. | A first-try pass rate with a 95% confidence interval, a rate per named behavior, and a before and after with an interval on the difference. |
| After the test | Engine writes a prompt or code change and opens a pull request. Datasets can be exported to fine-tune elsewhere. LangSmith does not train. | The failed runs become training data. Train with SFT, DPO or GRPO and host the result on your own endpoint. |
| Hosting | Deployment hosts the agent, with state, queues and autoscaling. It does not host models. | Hosts the trained model on an OpenAI-compatible endpoint. Your agent code stays where it is. |
| Customers | Expedia, Autodesk, Workday, Nvidia, ServiceNow, Coinbase, Monday.com, Lyft and others on its site. | Design partners in finance and customer support. Names not published yet. |
| Funding | $125 million Series B led by IVP, October 2025, at a $1.25 billion valuation. | Not disclosed. |
Pick LangChain if
- You build on LangChain or LangGraph and want tracing that understands them.
- You want one vendor for tracing, evals and hosting the agent itself.
- You want fixes as prompt and code pull requests, not a new model.
Pick While. if
- You want the fix to land in the model, with the weights yours.
- You have little or no traffic yet and still need a test.
- You want a pass rate with an interval, not a score in a table.
- You want the test to run offline, with no key, in seconds.
LangChain alternative
If you came here for a LangSmith alternative: While. covers judges, datasets cut from your traffic, regression tests and a before and after, in an open SDK that runs offline with no key. It does not host your agent, trace LangGraph internals, or edit your prompts for you. What it adds is the other kind of fix. When a prompt change is not enough, the failed runs become training data, the model is trained and served, and a held-out test says whether it worked.
Questions
- Is LangSmith open source?
- No. LangChain, LangGraph and Deep Agents are open source under MIT. LangSmith is a closed platform with a free tier, and self-hosting is an Enterprise add-on.
- Does LangSmith train models?
- No. Engine writes prompt and code changes from production failures and opens pull requests. Datasets can be exported to fine-tune with another provider. There is no training or model serving in LangSmith.
- Can I use LangSmith and While. together?
- Yes. LangSmith accepts and forwards OpenTelemetry traces, and While. reads OpenTelemetry. Keep LangSmith for tracing and hosting the agent, and use While. to train on the runs that failed.
Sources
Everything here about LangChain comes from these pages, read on September 17, 2026. If something changed, tell us and we will fix it.