LightfieldAI-native CRM that builds itself and does work for you
Promoted
Launching today
e-volv
AI agents across GitHub, GitLab, Jira and Linear
4 followers
AI agents across GitHub, GitLab, Jira and Linear
4 followers
Draw a workflow, and agents run it across your whole pipeline: review PRs, triage issues, break down tickets, write changelogs, gate releases. Works across GitHub, GitLab, Jira, Linear and ClickUp. Your model keys, your spend, human approval where you want it.
Hi Product Hunt 👋
A pull request opens in GitHub. The ticket is in Linear. The changelog is a file someone edits by hand. Nothing connects them, so a person does — reads the diff, finds the ticket, checks the acceptance criteria, moves the status, writes the line.
Evolve automates that whole path, not one step of it.
You build a workflow on a canvas: a trigger, a clone, an agent, an approval gate, a publish step. It fires on a webhook or a schedule. The agent reads the diff and the linked issue together, posts the review, moves the ticket, writes the changelog entry.
88 workflows ship prebuilt. Code review and PR triage, yes — but also: break a ticket into sub-tasks with acceptance criteria, generate a technical design before anyone writes code, detect breaking changes in an API contract and name the affected consumers, find flaky tests across recent CI runs, scan dependency licences, backport a merged PR on a label, sweep dead code. Import one, point it at a repo, run it.
Three things that shape how it works:
Every agent picks its own model. Claude on the security review, something cheap and fast on the changelog. Your keys — OpenAI, Anthropic, Azure, Bedrock, OpenRouter, DeepInfra — so the spend is yours at your rate. Your code goes to the provider you picked, under the agreement you already hold with them. We don't train on it; there's no training pipeline in the product to do it with.
You can stop it. Dry-run any workflow and every step that would write returns what it would have done instead of doing it. Approval gates hold a run until a named person decides. An agent only holds the tools you give it, so a review agent without write tools cannot push a commit whatever it concludes.
You can see what it did. Every run records each step's inputs, outputs, model, duration and token cost, and replays.
You can see what it did in production, too. Evolve Observer takes logs and traces from any service — Node, Python and Go SDKs, an edge client, or plain OpenTelemetry — groups errors by the file and line that raised them, and when a burst hits, draws the failure graph across services and writes the root cause. It is a tier of its own, included with Studio.
It runs across GitHub, GitLab, Bitbucket, Azure DevOps, Jira, Linear and ClickUp — which is the part Copilot, GitLab Duo and Rovo can't do, because each stops at its own platform.
Public beta.
Free tier: 1 workspace, 3 agents, 100 executions a month, plus 20 AI runs that need no API key at all. No card. Pro is $29/user with a 14-day trial.
I'll be here all day. Hard questions especially — particularly from anyone burned by a tool that wrote something confidently wrong.