Product at @langchain. prev product at @plotly Email me: nathan (at) drezner (dot) xyz
Montréal, Québec
- Tweets 886
- Following 1,968
- Followers 428
- Likes 17,177
Pinned Tweet
Browser use w/ @LangChain + @typesafeai's Jev! Really fun to build.
... I found it's excellent at playing the Wikipedia Game. (But, it's also great at "folding laundry" type tasks, like finding cheap flights.)
9
15
105
61,660
nathan drezner retweeted
Managed Deep Agents from @LangChain launched today with built-in web search powered by Parallel.
Add one MCP server to your tools file. LangSmith manages the credentials, runs the calls, and traces every search: no extra account or API key.
Bonus: It’s free during the public beta of Managed Deep Agents!
10
3
36
9,254
🔢 at @LangChain we've been putting a lot of thought into safe storage of user memories.
🔐 We wanted a secure way for a distributed agent to manage memory without it leaking outside of its intended context.
Here's a quick overview of the system and how to add user and agent memory to your managed deep agent - it's a single line of Python or TypeScript
7
3
16
2,545
nathan drezner retweeted
Parallel is now built into @LangChain managed agents.
Excited to partner with @p0 on Managed Deep Agents! You now have parallel built into your agents so your agents search the web fast and efficiently!
@travers00 @hwchase17 @VictorMoreira16
6
8
53
15,122
nathan drezner retweeted
Excited to partner with @p0 on Managed Deep Agents! You now have parallel built into your agents so your agents search the web fast and efficiently!
@travers00 @hwchase17 @VictorMoreira16
6
18
43
18,495
nathan drezner retweeted
𝚖𝚊𝚗𝚊𝚐𝚎𝚍-𝚍𝚎𝚎𝚙𝚊𝚐𝚎𝚗𝚝𝚜 0.8 is here and I think it is the best thing since sliced bread
things we shipped:
🧠 agent and user memory with access policies
🌐 connect webhooks with the new HTTP channel
✋ improved human-in-the-loop support
💬 Slack channels now support images and files
🔑 improved OAuth connection options
🔒 authenticate sandbox requests via a secure proxy
🔎 search the web with Parallel (thanks @p0)
📁 a runtime sandbox files API
released for js and py! langch.in/mda
14
12
44
4,837
Managed Deep Agents now supports user memories.
You can control where memories are stored, and memories that are created from private contexts are never shared to other users 🔐
Context Hub acts as the management layer to securely store agent memories and make general agent memories inspectable for developers, while keeping user memories private.
Very proud of the team for this feature - durable user memory is a very tricky problem and we're really excited about the direction we're taking here.
3
3
18
7,310
Get started using durable memory by checking out our documentation: docs.langchain.com/langsmith…
3
104
nathan drezner retweeted
Introducing Managed Deep Agents 0.8
✅ User-owned credentials + memory let agents work off correct permissions & user-specific context.
✅ HTTP channels for bringing agents into internal tools
✅ Native @slackhq file transfers
✅ Built-in web search via @p0
langchain.com/blog/langsmith…
10
12
56
9,574
Managed Deep Agents 0.8 launch today @ Interrupt! Check out the keynote if you haven't yet.
There's a ton of great features in this release that take MDAs to the next level for Enterprise -
- 🔐 Durable, secure user memory
- 🗄 Sandbox file API, so your agents can easily move files in and out of the sandbox
- 🌐 HTTP channels enable custom service triggers for your agent
- and more
I'll be sharing more over the course of the day and we have some great material to share about the feature set so watch this space
1
3
11
1,696
nathan drezner retweeted
some questions re context engineering for jev:
* what should jev do, what should jev not do?
* how to represent a problem as state and questions?
* how to present state and questions optimally?
* how can i calibrate next steps on jev's confidence / probabilities?
what else?
16
8
43
7,048
Take a dive on Schedules for MDAs with me! ⏰⏰⏰
You can add schedules to Managed Deep Agents.
1️⃣ Add a schedule file to schedules/ with a cron expression + a prompt
2️⃣ Deploy
...And the agent runs itself
💻 Docs: docs.langchain.com/langsmith…
⏯️ Demo from @ndrezn ⤵️
1
2
95
Build a coding agent in 5 minutes - meet 🤖 @.Patch
The Managed Deep Agent is only a few dozen lines of code, but it leverages Connections for Github, Slack channels, and a human-readable instructions.md for setup.
Such a good example of what's possible with MDAs!
With Managed Deep Agents, you can build agents that ship GitHub PRs from Slack with just a few lines of code.
We built one called Patch
💻 Tag it in a thread
💬 Describe the feature
…and it opens a PR with the diff, ready to review.
docs.langchain.com/langsmith…
1
117
nathan drezner retweeted
🖼️ Improved UI for Jev models in LangSmith
agents are going to complex systems of many llms calls
many llm calls will be jev-like (decision models)
we've updated tracing view in LangSmith to display these types of calls (state, questions, choices, output) more clearly!
41
9
90
21,014
I would be very curious to see the external evaluation. Is it public? I couldn't find anything
Opus 5.5 is our first model since we called for pacing the frontier. As with previous models, it was tested by external evaluators before release, including METR and Frontier Design.
On our most comprehensive alignment test, it achieves the strongest score to date.
76
If you're using managed deep agents in LangChain, take our @typesafeai Jev middleware for model routing for a spin.
Define your routes and Jev handles model selection.
3
4
18
11,731
nathan drezner retweeted
Polars 2.0.0rc2 has been released.
We expect this to be the last release candidate and that Polars 2.0 will go live next week.
Install now by running `pip install --pre polars`
Link to the full release guide: docs.pola.rs/releases/upgrad…
20
178
10,739
Great writing from colleagues on using Jev as Judge - cheaper, faster than similar tasks with chat LLMs. It's an excellent use case.
We tested Jev against LLM judges on accuracy, repeatability, latency, and cost to see whether System One models could offer a new approach to agent evaluation.
2
11
2,135
if you're curious you can see great examples using dev as a judge here: github.com/danielgshea/jev-a…
2
47