Back to a client after three months
Agencies and freelancersFreelancer with repeat clients
A client came back for phase two. The freelancer had forgotten the deployment quirks and why the cron job ran at 3:17.
How do I get my AI agent up to speed on a client project after three months away?
Open a new session in the client's repo and ArcRoom's recall hands the agent the project's decisions, rules and saved answers from phase one. Phase two starts where phase one ended, including quirks like why a cron job runs at 3:17. Expiry settings may have removed idle memories, so set longer idle days for client projects.
Also asked as: I forgot why a client's cron job runs at an odd time. How do I avoid that next time? · How do I restart an old freelance project without re-learning everything? · Can an AI agent remember a client's deployment quirks between projects?
Decisions, rules and saved answers are still in the project.
- Decision
- Recall
- Expiry rule
Decision (Jul): cron at 3:17 to avoid the payment gateway's 3:00 maintenance window.
Step by step
- Decisions, rules and saved answers are still in the project.
- The first session gets them in recall.
- Phase two starts where phase one ended.
What the agent sees
Decision (Jul): cron at 3:17 to avoid the payment gateway's 3:00 maintenance window.Compared with today's tools
Handover doc in Notion + CLAUDE.md
At the end of phase one the freelancer writes up deployment quirks and the reasons behind odd choices, then reads it before phase two.
- Setup
- 1–2 hours at wrap-up
- Ongoing
- only as good as notes written at the tired end of a project
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 2 min (npx arcroom init)
- Ongoing
- None
A good handover doc works if it gets written; ArcRoom keeps the decisions saved during the work and puts them in the first prompt back without anyone opening the doc.
Where it stops
Expiry settings may have removed idle memories. Set longer idle days for client projects.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.