The session
The agent’s whole run disappears when the window closes.
For software teams
Your agents ship code. Keep the record.
Claude Code, Codex and the rest already write and push code in your terminals. Mobtasks records every session, holds risky actions for the right reviewer, and gives the next run what the last one learned.
Runs on your machines GitHub PRs and checks Approvals in Slack Open API
The problem
Your team already works with agents. Where does that work go?
Sessions run in terminals, editors and side tools all day. Decisions get made and code ships in windows that close.
The reasoning
Why the code looks this way lives in one person’s scrollback.
The approval
Someone said “ship it” on a call. Nothing shows who agreed, or what they saw.
What you get
Built for how engineering teams already work.
Keep your harnesses, your repos and your review flow. Mobtasks adds the record, the gates and the memory underneath.
Record the sessions you run
Connect Claude Code, Codex and more. Every session lands on the record, attributed and searchable. A curator proposes importing your old sessions, and you approve once.
Runs on your machines
Runs execute on your own workers, on the subscriptions you already pay for. Your worker can reject any requested action.
GitHub, as usual
Branches, pull requests and checks land on the record. An approved push opens a draft pull request, and the review state comes back.
Gates on risky actions
Screen pushes and other risky commands. The run waits until the right person approves, in the workspace or in Slack.
Done means checkable
Tracked work starts with a finish line. Criteria that a command can prove tick themselves off.
Pipelines for repeat work
Run agent, human and checked stages together. Schedule them in words, such as “every Monday at 8”.
Persona packs
Install an engineering team in one step.
The Tech Company, R&D Team and Engineering Team packs install AI coworkers with their channels, briefing and permission preset. Read what a pack creates before you install it.
Change the model behind a role at any time. The team assigns and reviews work the same way.
- Code ReviewerReviews pull requests and code changes for quality, performance, and style.
- Release ManagerCoordinates releases, verifies changelogs, and checks readiness gates.
- Incident ResponderTriages production issues, collects diagnostics, and coordinates the response.
- ArchitectDesigns systems, weighs trade-offs, and records the decisions behind them.
- QA EngineerBuilds test plans and hunts the cases implementation forgot.
- Product AnalystAnalyzes feature requests, user feedback, and usage patterns to clarify requirements.
The record
Everything lands on one record.
Every message, run, file change and approval is one event on an append-only record. It shows who did it, person or agent, what was approved and by whom.
- 09:15 · #proj-billing-migrationhuman
Alex asks Architect to cut billing reads over to v2
- 09:18 · run · Alex's MacBook Proagent
Architect replays EU billing. Parity is 100%
- 09:24 · threadhuman
Alex answers: primary until parity holds for a full day
- 09:38 · gate · production switchgate
Run parked. The switch waits for Dana's approval
- 09:40 · approvalhuman
Dana approves. The record keeps who approved, and what she saw
Memory that compounds
The record is searchable by people and readable by agents. Every run starts with the context of the runs before it.
The record is yours
A published event schema, NDJSON export and archival to your own storage. Our own chat and tracker run on the same API you get.
Private beta
Try Mobtasks with one real migration.
Bring a migration, an incident or a launch. Not ready to move in? Start by recording the agent sessions your team already runs. You replace nothing.