01Missing context.
You reach a limit in Codex and switch to Claude Code. Your code may be available, but your decisions and next step aren’t shared automatically.
Example handover · Sign-in flow
Your AI hit a limit.
Your project shouldn’t stop.
Save decisions, progress and next steps in Memford. Continue in your next connected AI tool.
02Save before switching.
Before you need to switch, ask your connected agent to save the decision, what’s done and what’s next. Memford keeps that handover with your project.
Example saved handover: decision, GitHub sign-in; done, the login screen opens the sign-in flow; next, test an expired session.
03Connect the next tool.
Your next tool or AI account needs approved access to the same Memford project. Here, Claude Code connects to the handover Codex saved.
Check your tool04Continue from here.
Ask Claude Code to read the last saved state. It now has the decision, the progress and a concrete next task: test an expired session.
Keep the project clear
as it changes.
As work moves on, notes accumulate and decisions change. Keep your project context useful for the next session, the next tool and the next day.
Keep up with changes
while you work.
One agent finishes the expired-session test and saves the result. A supported connection can pick up that update at the next suitable step, so you don’t have to carry it into the other tool yourself.
Push updates & Streamable HTTP
A subscribed client receives a change hint, then loads the current context. Streamable HTTP carries these hints; the client still needs to subscribe and pass the loaded context to its agent.
Codex and Claude Code use approved local hooks to check at supported steps. An update does not interrupt an answer in progress. Other clients may need you to ask for a fresh read.
Set up automatic updates How subscriptions work- New result saved“Expired session returns to sign-in.”
- Connected client loads the updateAfter a change hint or a supported check.
- Context provided at the next supported stepThe receiving agent can use the new result.
Received, loaded and provided are separate steps. Delivery does not prove the agent used the information.
Less repetition.
A summary you approve.
After several sessions, the same decision may appear in many notes. With your permission, AI review prepares a shorter working context with references to its sources.
Review the draft, then approve or reject it. Connected agents can request the approved summary alongside notes it does not cover and complete assigned rules. The originals stay available.
AI can miss details. Optional automatic drafts prepare the next review; they never approve it for you. Changing a covered source invalidates its approved summary.
How summaries are reviewed“Use GitHub for sign-in.”
“Login screen opens the sign-in flow.”
“Expired-session test passed.”
Draft for your review
Use GitHub sign-in. The login flow opens, and the expired-session test passed.
Sources: decision, implementation and test notesOnly a valid, approved draft is available as compact context.
When decisions disagree,
check the sources.
One note says GitHub sign-in. Another proposes email links. Use AI review for decision triage: inspect reported facts, decisions, hypotheses and possible conflicts alongside the notes they came from.
You decide what should stand. Correct the source notes or add the agreed outcome, then generate and review a fresh draft.
AI may miss a conflict or flag one incorrectly. It does not choose the project’s direction or silently turn a proposal into an approved decision.
Review conflicting decisions- Saved decision
- “Use GitHub sign-in.”
- Later proposal
- “Switch to email links.”
Was the decision changed,
or is this still a proposal?
Check both sources. Record the intended outcome before approving a new summary.
Change your mind.
Keep the history.
Decisions evolve. Version history keeps the original submission and later edits or status changes, with their revision, receipt time and recorded account. You can see how a note changed without losing what came before.
Version checks before saving
A protected save also checks whether the shared context changed while the agent was working. If it did, the agent must read and reconsider the update before saving a new contribution.
History does not roll back a running agent or its files. Older connections may save without a version check; the check protects the Memford contribution.
Explore version history- Original
“Test an expired session.”
- Later revision
“Expired-session test passed. Next: check sign-out.”
The original stays in the history. Active reads use the current version of the note.
Your ways of working can travel with the context, too. Share rules across projects and let agents discover reusable skills.
Start free.
Connect your first AI tool in three steps.
- Copy and paste the prompt into your AI tool.
- Sign in and choose your project.
- Approve the connection.
Use the same project for each tool. Keep passwords and keys out of chat.
Read the setup guideFor updates while you work, enable automatic updates.
A few things to know.
Can I switch AI tools or accounts?
Yes, when your next supported AI connection has approved access to the same Memford project. Codex and Claude Code are supported examples; other tools vary. Your provider’s conversations and account state do not move. Check your tool.
Doesn’t GitHub already keep the project?
GitHub keeps what you commit. Chat decisions and next steps aren’t included automatically. Memford makes the knowledge you save available to your next connected agent. Files, worktrees and running processes stay where they are. Code access remains separate. About shared context.
What if I hit a limit before saving?
Your next agent can read only what was saved. Memford does not recover an unsaved conversation. Save a useful handover while you work, before you need to switch.
Can changes arrive while I’m working?
Yes, with automatic updates enabled in a supported tool. An update can arrive with your next message or after a tool finishes, at a supported step. It does not interrupt an answer being written. Codex and Claude Code need the current local connection helper and your approval; other clients read when asked. Large context may arrive as an attachment the agent needs to open. Enable automatic updates or update your setup.
Can I share rules and ways of working?
Yes. Memford supplies assigned project rules in full; agents can discover reusable procedures and load the full instructions when needed. Optional AI summaries remain drafts until you review and approve them. Originals and revisions stay available. Explore the features.
What stays under my control?
You approve each connection and can revoke its project access. A protected Memford save checks whether shared context changed before adding your note. If it did, the agent must review the update first. Retrying that exact save does not create a duplicate. Older connections can still save without this version check; other file edits and tool actions are outside it. About protected saves.