Reading One Piece. Slow gains. New art form every year (2026: TBD). Bass under construction. Owned by one cat. Marcus Aurelius was just a guy with a journal.
We’ve added custom classifiers to Tessary. You can now describe a failure specific to your agent and have Tessary build a classifier to monitor it in production.
Tessary is now available on Cloud. You can try our open-source agent reliability platform without self-hosting.
Tessary monitors every production trace, detects issues that sampled evals miss, and investigates to find the root cause.
We’ve added a frustration classifier to Tessary.
It finds conversations where users are unhappy with your agent, detects when frustration increases, and helps identify what is causing it.
A tool shared by 3 of our agents went from 1.7% to 14.8% failed calls in a week. We missed it because their tasks were still getting done.
Tessary caught the increase, traced the failures to how the agents were using the tool, and suggested clearer tool definitions. Link Below.
New release: Repository connection is now faster and easier, helping RCA attribute issues back to changes in your code or prompts.
We’ve also improved triage accuracy, fixed several bugs, and added automatic model price updates for self-hosted instances.
https://t.co/6h8JVdYVnl
Today, we’re launching Tessary, an open-source agent reliability platform.
Tessary monitors every production trace, detects issues that sampled evals miss, and investigates to find the root cause.
Agent of agents: Layered agents that simplify tasks as they go
Adaptive agents: Agents that learn how to orchestrate and co-ordinate, the ML equivalent really
Why do I get the feeling that the progression of how people deal with agentic coding is following basically how software engineering evolved?
prompting > vibing > looping > graphs
Isn't that reminiscent of complexity of data pipelines?
If so, logically the next ones in this progression would be:
@thenowhereway@dilip_designs Most threads on Reddit have started to feel like a sales pitch though, how are you guys still managing to find the ones that convert meaningfully?
Happy to report that our crew of focussed agents have continued to run, learn and ship code over the past 3 months, as our code base continues to grow. We've open sourced the "crew" here:
https://t.co/imt65Ldmsu
Pro tip: Ask the crew to trigger adverserial checks for big diffs.
Writing a DnD campaign feels easier than prompting agents tbh. At least your players get punished immediately for not checking in before they hallucinate.
@addyosmani I wonder if there is a way around this by bringing more predictability to the way these coding agents think and solve the problems they are assigned. Not being able to keep up with the agents is a reality, but may be we can surface the few that do need constant attention?
@skeptrune@greptile Doesn't its pricing plan discourage PRs at the rate that teams are shipping them?
Our repos cross the included limit within a day, and choosing to author larger PRs or letting go of stacks seems to be the poorer choice.
Kosho has a level of simplicity that none of the other worktree tools seem to match.
Others I’ve tried all add another layer to learn. Kosho stays out of the way and works naturally with agentic harnesses, CI pipelines, and terminal workflows. I really don’t need anything else.