Augment Code Fixes Chat Saving and Agent Edit Failures in VS Code
Version 0.904.1 changes how failed conversation saves are retried and removes edge cases that could break large histories or agent file edits.
Edited by Tyronne Panaino
Augment Code released VS Code extension 0.904.1 on October 7 with a set of reliability fixes for chat-history persistence and coding-agent edits. The update changes how failed saves are isolated and retried, prevents very large histories from failing on a call-stack error, and repairs an edit path that could break when another language extension supplied an incomplete diagnostic record.
The release matters to developers who depend on an AI coding session remaining recoverable across long conversations or intermittent storage failures. It also narrows a specific integration failure at the boundary between Augment and VS Code language tooling. The evidence is limited to Augment's own release notes, so the fixes are confirmed as shipped changes but not independently measured across different workspaces or extensions.
Failed saves stop blocking newer conversations
The central persistence change separates one troublesome conversation from later save work. Augment says a conversation that repeatedly fails to save no longer prevents newer conversations from being written, and the extension no longer marks that failed conversation as saved before a write actually succeeds.
That distinction is operationally important. A saved indicator should represent completed persistence rather than the intention to persist. By allowing later conversations to proceed, the extension also reduces the chance that one recurring failure leaves an expanding queue of newer work waiting behind it.
Version 0.904.1 adds automatic recovery for that queue. A large backlog of unsaved conversations can now drain without requiring another chat action, while transient failures are retried with backoff. The release notes do not define the retry schedule, storage implementation or terminal failure condition, so users still lack enough public detail to predict how long recovery should take in a particular environment.
Large histories get a separate loading fix
Augment also says very large chat histories no longer abort during saving or loading when the extension encounters the `Maximum call stack size exceeded` condition. This is presented as a fix for the failure mode, not as a new capacity guarantee. The notes do not publish a conversation-size limit, a test corpus or memory measurements.
A related cleanup limits conversation-title generation to one request per conversation. Augment attributes the earlier duplicate behavior to overlapping triggers. Removing those duplicate title requests should reduce redundant work inside the extension, although the source does not quantify latency or resource savings.
Agent edits now tolerate incomplete diagnostics
The coding-agent fix targets information received from other language tooling. Augment says agent file edits and diagnostic lookups no longer fail when a language extension reports a diagnostic without an error code. That is a narrow interoperability change: the agent path now tolerates the missing field instead of allowing that diagnostic shape to break the operation.
The release notes do not name the language extensions that produced the condition, say how often it occurred or describe the resulting edit state after an earlier failure. Teams that saw unexplained edit failures should therefore treat 0.904.1 as a relevant remediation checkpoint, not as proof that every edit or diagnostics integration problem has been eliminated.
The same version removes Augment Migration commands from the VS Code Command Palette when migration is unavailable to the user. That prevents an unavailable workflow from being presented as an actionable command, but Augment does not specify which account, project or rollout conditions determine availability.
What teams should verify
After updating, teams with long-running chats can check whether previously problematic histories save and reopen, whether newer conversations continue saving after one failure, and whether any backlog clears without a new chat action. Reproducing the diagnostic edge case requires a language extension that emits a diagnostic without an error code; the public notes do not provide a test fixture.
The next useful evidence would be telemetry showing reduced save failures, clearer documentation for retry exhaustion and compatibility results across common language extensions. Until then, this is a concrete maintenance release with a clearly described failure boundary, supported by one first-party source rather than an independent reliability evaluation.
Status
Confirmed. Augment Code published the dated 0.904.1 release notes and describes the changes as shipped fixes. Internal confidence is medium because the evidence is first-party only and includes no affected-user count, independent test results, rollout measurement or performance data.
Sources
Update note: Last reviewed 2026-10-13. We will revise this post if Augment publishes compatibility details, recovery limits or measured reliability results.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.