--- {"title": "How to compare agent runs across branches in 2026", "description": "Compare agent runs across branches step by step, with the details to record, checks to make, and tools that can help with session views and cost.", "url": "https://supakeeper.com/how-to-compare-agent-runs-across-branches-for-ai-tooling-researchers/", "date": "2026-11-03"} --- Compare agent runs across branches step by step, with the details to record, checks to make, and tools that can help with session views and cost. To compare agent runs across branches, collect the runs, record the same details for each one, and compare runs of the same task side by side. AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [^1] | Option | What it can do | What to check | |---|---|---| | AgentHydra | AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [^1] | Check whether each session shows the branch and task details you need. | | Existing agent tooling | Check what the existing workflow provides for comparing agent runs. | Check whether it shows the runs and details needed across the branches being compared. | | Separate session tool | Check what the separate session tool provides for comparing agent runs. | Check which agents and run details it supports before choosing it. | ## What you need before you start Before you compare agent runs across branches, have the runs you care about still visible on your machine. Know which branch and which task each run belongs to before you open anything, because runs from several branches are easy to mix up. Write down the question you are answering, such as which setup gave the better result or which branch needed fewer steps. A clear question tells you which details to compare and which details can be ignored. Have one consistent set of details ready for every run: the branch name, the model, the date, the task description, and the outcome. Runs that record these the same way are easier to line up side by side. Decide ahead of time where you will note results, whether in a notes file, a spreadsheet, or a simple table. Check which agent tools produced the runs. If your work uses more than one agent, confirm that the tool you use for the comparison can show the sessions you need. If you plan to look at cost, decide which price basis you will use, since token usage and billed amounts can differ. ## Step by step 1. Collect the runs from each branch into one place. Include the branch each run came from and the point where it started and ended. 2. Give every run the same label pattern, such as branch, task, and date. Consistent labels make later searches faster. 3. Open each run and record its outcome against the question you wrote down. Note only what you can see in the run itself. 4. Compare runs of the same task only. Runs on different tasks rarely answer the same question, so a comparison between them can mislead. 5. Check the conditions for each run, including the model version, the prompt or instructions used, and any files the agent changed. 6. Search the session list when looking for a specific run, and check whether the session details include the branch or task information needed for the comparison. Keep the comparison set small enough that you can read every run in it. 7. Write down what differed and what stayed the same. Then decide whether the difference comes from the branch itself or from the setup around it. 8. Repeat the comparison after a rerun. A single run can vary, so a second look helps confirm a pattern. ## Tools that can do it AgentHydra is made by the team that publishes this site. AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [^1] Check whether the session view shows the branch and task details you need. AgentHydra shows cost at API list prices by day, model, project and account. [^1] The same cost numbers are printed as JSON from the command line. [^1] Read these numbers as list-price figures when you compare runs. AgentHydra pairs each failed shell command with the command that fixed it, and can copy those pairs out as a rules file for your agents. [^1] AgentHydra is free, with no account or subscription. [^1] It is open source under the MIT License. [^1] On Windows, install it with a one-line PowerShell script, download a build from GitHub Releases, or run it from source with Bun. [^1] The tray app needs Windows; macOS and Linux builds exist but have not been verified. [^1] Other approaches include the logs and session views already available in an agent workflow, or a separate session tool. Check each option against the exact job of comparing runs across branches, including the agents it can show and the run details it records. ## Common mistakes to avoid Avoid comparing runs from different tasks as if they were the same test. A difference in outcome may come from the task rather than the branch. Avoid comparing runs made with different model setups without noting the difference. Record the setup for each run so the comparison stays fair. Avoid treating cost figures as the amount you were billed. Check which price basis the figures use before you draw conclusions about spending. Avoid using a single run as the basis for a decision. One run can vary, so repeat the comparison before you act on it. Avoid unclear branch names or missing task descriptions. A list is harder to use when runs do not show enough context to tell them apart. ## Frequently asked questions ### How can applied AI researchers compare agent runs across branches? Start by collecting the runs from each branch into one place and recording the same details for each one, such as branch, model, task, and outcome. Compare only runs of the same task, and check the conditions of each run before drawing a conclusion. Repeat the comparison after a rerun, because a single run can vary. A searchable session list can make the collecting and searching steps quicker. ### Can AgentHydra show agent runs from several branches in one list? AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [^1] Whether each session shows its branch depends on what the session records, so check that in your own sessions before relying on the list for this comparison. Search the list using the details the sessions actually record. ### Is AgentHydra free for comparing agent runs? AgentHydra is free, with no account or subscription. [^1] It is open source under the MIT License. [^1] On Windows, install it with a one-line PowerShell script, download a build from GitHub Releases, or run it from source with Bun. [^1] The tray app needs Windows; macOS and Linux builds exist but have not been verified. [^1] ### Does AgentHydra show the cost of each agent run? AgentHydra shows cost at API list prices by day, model, project and account. [^1] The figures use API list prices, so check that price basis before drawing conclusions about spending. The same cost numbers are printed as JSON from the command line. [^1] Check in the session view whether branch details appear alongside the information needed for your comparison. ### What should be checked before choosing a tool for comparing agent runs? Check which agents the tool reads, since the runs may come from more than one. Check whether each run shows the branch, task, model, and date needed for the comparison. Check how the tool searches its run information, and whether it updates while runs are still going. Finally, check the platform and price terms, and confirm that the tool fits the machine before using it for comparisons. Free, with no account or subscription. [^1] - [Visit AgentHydra](/go/start) ## Sources [^1]: AgentHydra's own description of itself (the product of this site's publisher) (/go/start)