InsForge
The agent-native cloud infrastructure platform
1.8K followers
The agent-native cloud infrastructure platform
1.8K followers
InsForge is the agent-native alternative to AWS.
This is the 3rd launch from InsForge. View more
InsForge Backend Branching
Launching today
Git-style branching for your entire backend. Spin off a branch to get a full, isolated copy of everything: your database, storage, auth, edge functions, and more, ready for your agents to experiment on without touching production.
Launch Team / Built With
congrats on the launch!
how does conflict detection work when two branches modify the same backend resources?does the agent receive guidance on resolving them?
InsForge
@emily_carter18 yes. We compare parent at branch creation, parent now, and branch now. If the same table / policy / function / config diverges, we block the merge and show a conflict report. The agent can use that context to fix the branch, then retry.
curious about performance.does branch creation duplicate data physically or use snapshots under the hood?how does that impact costs?
InsForge
@joshua_cooper2 We support two branch modes: `full` copies schema + data at branch creation time, while `schema-only` copies schema and mergeable backend config but leaves user-data tables empty. For huge prod DBs, schema-only is usually the lighter path, then you add seed/test data as needed.
How detailed is the audit trail? can teams see exactly which agent initiated a change?that could be valuable for accountability.
InsForge
@luz_bidelspach totally agree, accountability matters a lot once agents can touch backend state.
Today the branch / diff / merge flow gives teams visibility into what changed before merge. More detailed agent-level attribution is something we care about, but I don’t want to overclaim it as fully solved yet.
Branching for backends is wildly underrated — we just hit the same pain when migrating our backend from Python to Cloudflare Workers (different infra, same need to test risky changes in isolation). Curious: do you handle data branching too, or just code/config? Either way — congrats on shipping 🚀
InsForge
@mythic_dd thanks! We handle both, with clear boundaries. In `full` mode, the branch copies schema + data at T0 for realistic testing. In `schema-only`, user-data tables start empty. On merge, schema and mergeable backend config can go back to parent, but user-data rows are not auto-merged into prod.
Really impressive concept. Quick question for the team how do you handle branch merging when there are schema migrations involved? That's always been the tricky edge case with database branching. Would love to understand your approach.
InsForge
@daniel_juan2 We use a 3-way diff: parent at branch creation, parent now, and branch now. If the branch migration changes a schema object that parent didn’t touch, it can be previewed and merged. If both sides changed the same table / policy / function differently, merge is blocked with conflict context instead of guessing.
Does the merge diff separate schema changes from data changes and config changes? when I am reviewing what an agent did, those are really three different risk levels that need different kinds of attention. would love to know if the review UI reflects that layering or surfaces everything together.
InsForge
@carter_garcia yes. Merge review focuses on mergeable backend objects: schema, RLS, functions, auth/storage config, realtime, schedules, etc. User-data rows are isolated and not auto-merged into prod, so data changes don’t get mixed into the same risk bucket.
Congrats on the launch! Can you plz tell how does the semantic layer handle it when the agent makes a wrong assumption about the schema? does it self correct or does it need manual intervention?
InsForge
@munis_abbas We don’t rely on the system “guessing” the right schema semantics. The branch gives the agent a safe backend to try the change, then merge uses structured diff + conflict checks. If the agent made a wrong assumption, you can review the diff, let the agent fix it on the branch, or reset the branch back to T0 and retry.