The agent just starts.
Without a workflow, the agent implements the requirement as it stands in the backlog. It does not question it or work it out further, and nobody reviews in between. The result often differs from what was meant.
A team of agents takes your requirement from use case to pull request, and your team decides and reviews. ollayr gives you control on every layer of that work: the brief, the agents and skills, the workflow and the token use. It works with Claude Code or opencode and with Jira, GitHub Issues, Figma and Zeplin.
story pulled in
acceptance criteria drafted
design linked
endpoints linked
tool call approved
Without a workflow, the agent implements the requirement as it stands in the backlog. It does not question it or work it out further, and nobody reviews in between. The result often differs from what was meant.
It works well for one developer and less well for the next, because each starts the agent with a slightly different prompt. How good the code is depends on who is at the keyboard.
The agent follows no fixed sequence. After each step it waits until someone tells it the next one, and whether tests and a review happen depends on who remembers.
Why the code looks the way it does, which commands the agent ran and who allowed them is at best in the history of a single session. Once that session is closed, nothing remains that explains the decisions behind the code.
ollayr installs an agentic workflow into your project folder. One agent writes the use case, one the solution concept and one the implementation plan. Then come implementation, tests and code review, and the workflow ends with a pull request. Reviewer agents check the documents and the code before a person does. You place an approval step wherever someone should decide and remove it where reviewing the result afterwards is enough. The work can start before the backlog: from a project description the agent drafts epics, and you create the ones you want in Jira.
Epics, stories and bugs come from Jira or GitHub Issues, and production errors come from Sentry. You link designs from Figma or Zeplin and endpoints from your OpenAPI spec.
AI reviews the requirement, improves the description and drafts acceptance criteria and test cases. The prompt templates your team saved for this area are available in the assistant panel.
Before the handoff you see exactly what the agent will receive, including your guidelines and architecture notes. You pick a workflow and the agent runs on your machine. It asks before tool calls and stops at every approval step the workflow defines.
The workflow writes a use case, a solution concept and an implementation plan into your project, and depending on the workflow a test plan. The timeline shows your team every handoff, every approval and every open question, with the cost per requirement.
Requirements
Add CSV export to the order list
Users can export the filtered order list as a CSV file.
Acceptance criteria
The export contains the columns shown in the list
Active filters apply to the export
An empty list produces a file with headers only
AI reviewREADY
The ticket had no acceptance criteria. I drafted three for you to review.
Story links
Add CSV export to the order list
Zeplin designOrder list with export dialogLinked
EndpointGET /orders/exportLinked
Brief preview
What the agent will receive
TicketDEMO-142, description and acceptance criteria
DesignOrder list with export dialog
EndpointGET /orders/export
You can remove any part before the handoff.
Pending approval
Run a command
Claude Code asks to run this command on your machine.
npm test -- orders
Approve call by call or for the whole session.
Timeline
Timeline
Cost
Implement storyClaude Code$1.12 · 268k tokens
Implement storyopencode$0.72 · 144k tokens
Demo data. Cost and tokens are shown as far as the agent reports them.
Requirement, designs, endpoints, acceptance criteria, guidelines and architecture notes reach the agent as one brief. You can remove any part, and ollayr never shortens it without telling you.
ollayr reads the agents, skills and MCP servers from your project folder. Your team edits them in one place and writes its own skills or adds an established one by pasting its GitHub link, for example Matt Pocock's grill-me skill. Prompts live in a shared library.
The workflow sets which agent does what, in which order, and where a person approves. You change it on a canvas or build your own, and the agent follows it.
The dashboard shows cost and tokens per requirement, workflow and agent, as far as the agent reports them. You see which workflow, which agent and which feature used the most.
Traceability comes from two places: the documents the workflow writes into your repository, and the timeline. The timeline lists each handoff with the person, the agent and the workflow, plus every approval and every open question from running sessions. The dashboard adds up cost and tokens per requirement, workflow and agent, as far as the agent reports them. You can compare workflows and agents side by side.
TIMELINE
COST
Available integrations are labeled above. Planned connections are not yet supported.
You hand over a requirement and get back a use case to approve, then a plan, then a pull request. You work in ollayr, without a terminal or prompt craft.
Your developers decide and review, and the agents write the documents, the code and the tests. Every approval and the cost are recorded per requirement.
Every client project starts with the same workflow and the same agents, however the team is staffed. You can show the client the documents, the approvals and the cost per requirement.
With Claude Agent for Jira, the agent works in Anthropic's cloud sandbox. With ollayr, it works on your machine.
With Jira, you review the result once it is finished. With ollayr, you approve the steps you chose before they happen.
Jira knows the ticket. ollayr also knows the design, the API spec, the acceptance criteria and your guidelines, and shows you what the agent gets before the handoff.
Yes, an agent expert can build most of it, and for one person working alone ollayr is not worth it.
ollayr turns such a setup into a team tool: product owners prepare briefs without learning agent engineering, and the workflow, skills and MCP servers are installed and updated from ollayr.
The team also sees what every agent did, who allowed it and what it cost.
The coding agent works on your machine through the ollayr local bridge, and your repository is not uploaded to ollayr. ollayr stores the brief text, the working directory and, for the timeline, file paths, tool titles, error messages from runs and the start of follow-up briefs on Appwrite Cloud in Frankfurt. What the agent sends to its own AI provider is covered by your agreement with that provider.
Claude Code and opencode work with ollayr today, and GitHub Copilot CLI is next. ollayr connects to agents through the open Agent Client Protocol (ACP).
You can connect Jira and GitHub Issues for requirements, Figma and Zeplin for designs, OpenAPI or Swagger specs for endpoints, and Sentry for production errors. A skill syncs the documents of a feature to Confluence.
You decide that per workflow. The workflow stops wherever you placed an approval step, and it runs through where you removed one. During a run the agent asks before tool calls, and depending on the agent you can allow a tool for the whole session.
All members of the project can, in the timeline. It shows who handed off, which agent ran with which workflow, which approvals were given and which questions are open.
AI usage runs on your own key or on your coding agent's login, so your provider bills it. ollayr shows cost and tokens per handoff, requirement, workflow and agent, as far as the agent reports them.
Tell us about your team and the requirements you want to hand off. We set up access and help you install the local bridge.
ContactYou need Claude Code or opencode on your machine and a project in Jira or GitHub. AI usage runs on your own key.