Code got cheap. Deciding what to build didn't.
Olympia runs from strategy to accepted PR, connected to your Jira or GitHub Issues and to your repository on GitHub. The AI reads the code, estimates and writes the cards, ready for the agent you already use. Your team decides the order, approves, merges and accepts.
- Strategy
- Priority
- Breakdown
- Refinement
- Development
- Acceptance
Why did the bottleneck move?
The bottleneck moved upstream of the code
Taking a card and opening the PR has become a feature of the tools the team already pays for. The expensive part is what comes before: what matters this quarter, what goes first, what done means.
A vague card becomes an almost-right PR. And an almost-right PR comes back.
of developers have run into AI solutions that are “almost right, but not quite”, the most cited frustration.
count project goals or requirements among the most important context to have when working on a task, ahead of code and documentation (73%).
lose 10 hours or more a week to organizational inefficiencies. In the 2024 report, it was 33%.
AI amplifies the strengths and the weaknesses a team already has.
Why not the Jira Product Discovery, Copilot or Rovo I already have?
Each tool you have covers one end
At one end is the roadmap. In Jira Product Discovery, impact and effort are fields on the idea, and a formula sorts the list. The effort is the number someone filled in.
At the other end is the PR. The Copilot coding agent takes the Jira card, asks its questions in Jira itself and opens a draft PR. Atlassian's Jira Coding Agent reads the card and the code, proposes a plan and opens the PR. Rovo also breaks epics into stories.
They are good tools. Both agents start from a card someone already wrote.
Jira Product Discovery, Productboard, Aha! Roadmaps, airfocus, Craft.ioJira Product DiscoveryProductboardAha! RoadmapsairfocusCraft.io
effort as a field, without reading the code
Rovo, Copilot, GitLab Duo, ClickUp BrainRovoCopilotGitLab DuoClickUp Brain
splits the big card into smaller cards
starts from a card someone wrote
Copilot coding agent, Jira Coding Agent, Devin, Cursor, OpenHands, Claude Code, JulesCopilot coding agentJira Coding AgentDevinCursorOpenHandsClaude CodeJules
takes the card and opens the PR
Strategy
Priority
Breakdown
Refinement
Development
Acceptance
Olympia
effort and risk read from the code, with the files that drove them
asks only what is the team's to decide and writes how to verify it
Olympia
Jira Product Discovery, Productboard, Aha! Roadmaps, airfocus, Craft.ioJira Product DiscoveryProductboardAha! RoadmapsairfocusCraft.io
effort as a field, without reading the code
Strategy
Priority
effort and risk read from the code, with the files that drove them
Rovo, Copilot, GitLab Duo, ClickUp BrainRovoCopilotGitLab DuoClickUp Brain
splits the big card into smaller cards
Breakdown
Copilot coding agent, Jira Coding Agent, Devin, Cursor, OpenHands, Claude Code, JulesCopilot coding agentJira Coding AgentDevinCursorOpenHandsClaude CodeJules
starts from a card someone wrote and opens the PR
Refinement
asks only what is the team's to decide and writes how to verify it
Development
Acceptance
Olympia connects the two ends. In the matrix, effort and risk come from reading the code, with the files that drove them. Between the matrix and the PR, Refinement asks only what is the team's to decide and writes criteria that say how to verify them. The refined card goes to your Jira or GitHub Issues, and the agent you already use can take it like any other card. Ideas, objectives and the matrix stay in Olympia, which does not read Jira Product Discovery or the other roadmap tools.
How does it work?
Six stages, six decisions your team makes
- Strategy
The team writes up to three objectives for the quarter, each with a number and a date. The AI points out what serves none of them.
Decision: What matters
- Priority
The AI suggests the value and reads the code to estimate effort and risk, and shows why. The matrix becomes the roadmap: Now, Next, Later.
Decision: The order
- Breakdown
The idea becomes epics, stories and tasks on the board.
Decision: What goes in
- Refinement
The AI reads the code and asks only the product questions, with a recommendation. Then it writes the acceptance criteria, each one saying how to verify it.
Decision: The answers and the approval
- Development
The refined card goes to the agent you already use, or to a person on the team.
Decision: The merge
- Acceptance
A person checks the merged PR against the criteria. Whoever sends it back says what is missing, and the card returns to Refinement with the reason.
Decision: Accept or return
Nothing moves forward without a decision, and each decision goes out under the account of the person who made it, in Jira and in GitHub. A card without at least one criterion that says how to verify it cannot be approved.
What changes in my backlog?
Every estimate says where it came from
Value comes from the objectives. Cost comes from the code. Each point on the matrix opens to show why, with the files that drove it.
When deciding whether to trust a technical answer from AI, 93% of developers say the source matters.
Is it safe?
Where each part runs and who sees what
In Olympia's cloud, Oracle Cloud in São Paulo
Strategy, Priority, Breakdown, Refinement, Acceptance
- cards, under the account of whoever decided
- ideas, cards, answers and code excerpts
- a copy of the code
In the services you connect
- The app and the database
- Oracle Cloud, in São Paulo, Brazil.
- Olympia's cloud AI
- Models from an AI provider contracted by Olympia, today Anthropic. They receive the ideas, the cards, the team's answers and the code they read for each card: the file list, the searches and the excerpts they open.
- The code, in Priority and Refinement
- A copy downloaded from GitHub to Olympia's cloud and deleted within 10 minutes of the last read. What the AI noted from it (facts with file and line, and the list of what it read) is kept with the card.
- Secrets
.env,.pem,.key,.p12andid_rsafiles stay out of the AI's reach. A key written inside the code does not.- Accounts
- Each person acts with their own, in Jira and in GitHub, and the GitHub app reaches only what that person can reach. Tokens and keys are stored encrypted.
- App usage
- PostHog and Microsoft Clarity record sessions. What people type is masked; the rest of the screen, cards and questions included, is recorded as it appears.
- Personal data
- Melo Tech, based in São Paulo, Brazil, operates Olympia and is the data controller under the LGPD, Brazil's data protection law. The privacy policy says what is kept, with whom and for how long. The other services that receive data, among them Anthropic, Cloudflare, Resend, PostHog, Microsoft Clarity, Jira and GitHub, are mostly in the United States.
Who uses it?
Olympia is built with Olympia
Olympia's backlog runs through Olympia, on the OLY board. Refinement reads the code, the agent works on one of the team's Macs, and a person merges and accepts.
How do I start?
Start by requesting access
For teams of 3 to 20 devs with their code on GitHub.
In Olympia's cloud
Every access request goes through approval. No charge during early access, with a monthly AI ceiling per team. When a team hits the ceiling, AI calls stop until the next month. Before any charge, each team gets 30 days' notice.
- Strategy
- Priority
- Breakdown
- Refinement
- Developmentwith the agent you already use
- Acceptance
Using Linear? Say so in your access request. The number of teams that ask decides whether the integration gets built.
What the team needs
The team needs a board in Jira or GitHub Issues and the repository on GitHub. The refined card is ready for the agent you already use.