Configuration - JoelBondoux/AtlasMind GitHub Wiki
Configuration
All settings are prefixed with atlasmind. and can be configured via VS Code Settings (Ctrl+,) or the AtlasMind Settings panel (AtlasMind: Open Settings).
Model Routing
| Setting |
Type |
Default |
Description |
atlasmind.budgetMode |
enum |
balanced |
Budget preference for model selection. Options: cheap, balanced, expensive, auto |
atlasmind.speedMode |
enum |
balanced |
Speed preference for model selection. Options: fast, balanced, considered, auto |
See Model Routing for details on how these settings affect model selection.
Tool Approval & Safety
| Setting |
Type |
Default |
Description |
atlasmind.toolApprovalMode |
enum |
ask-on-write |
When to request user approval before running tools. Options: always-ask, ask-on-write, ask-on-external, allow-safe-readonly |
atlasmind.allowTerminalWrite |
boolean |
false |
Allow write-capable terminal commands (installs, builds) after explicit approval |
See Tool Execution for the full approval and safety model.
Post-Write Verification
| Setting |
Type |
Default |
Description |
atlasmind.autoVerifyAfterWrite |
boolean |
true |
Run verification scripts after successful write operations |
atlasmind.autoVerifyScripts |
string[] |
["test"] |
Package scripts to run (e.g. ["test", "lint"]). Names are sanitised. |
atlasmind.autoVerifyTimeoutMs |
number |
120000 |
Max time (ms) for each verification script. Minimum: 5000 |
Chat Session
| Setting |
Type |
Default |
Description |
atlasmind.chatSessionTurnLimit |
number |
6 |
How many recent turns are carried forward as context. Minimum: 1 |
atlasmind.chatSessionContextChars |
number |
2500 |
Max characters for compacted session context. Minimum: 400 |
Memory (SSOT)
| Setting |
Type |
Default |
Description |
atlasmind.ssotPath |
string |
project_memory |
Relative path to the SSOT memory folder |
See Memory System for folder structure and retrieval details.
Project Planner
| Setting |
Type |
Default |
Description |
atlasmind.projectApprovalFileThreshold |
number |
12 |
Estimated changed-file count that triggers approval gating. Minimum: 1 |
atlasmind.projectEstimatedFilesPerSubtask |
number |
2 |
Heuristic multiplier for file impact estimation. Minimum: 1 |
atlasmind.projectChangedFileReferenceLimit |
number |
5 |
Max clickable file references shown after /project runs. Minimum: 1 |
atlasmind.projectRunReportFolder |
string |
project_memory/operations |
Folder for persisted run summary JSON reports |
See Project Planner for the full planning and execution flow.
Tool Webhooks
| Setting |
Type |
Default |
Description |
atlasmind.toolWebhookEnabled |
boolean |
false |
Enable outbound webhook delivery for tool events |
atlasmind.toolWebhookUrl |
string |
"" |
HTTPS endpoint for webhook payloads |
atlasmind.toolWebhookTimeoutMs |
number |
5000 |
Webhook request timeout (ms). Minimum: 1000 |
atlasmind.toolWebhookEvents |
string[] |
["tool.started", "tool.completed", "tool.failed"] |
Events to emit. Options: tool.started, tool.completed, tool.failed, tool.test |
Voice
| Setting |
Type |
Default |
Description |
atlasmind.voice.ttsEnabled |
boolean |
false |
Auto-speak freeform responses via TTS |
atlasmind.voice.sttEnabled |
boolean |
false |
Show voice input button (requires microphone) |
atlasmind.voice.rate |
number |
1.0 |
Speech rate (0.5–2.0) |
atlasmind.voice.pitch |
number |
1.0 |
Speech pitch (0–2.0) |
atlasmind.voice.volume |
number |
1.0 |
Speech volume (0–1.0) |
atlasmind.voice.language |
string |
"" |
BCP 47 language tag (e.g. en-US, fr-FR). Empty = OS default |
Budget
| Setting |
Type |
Default |
Description |
atlasmind.dailyCostLimitUsd |
number |
0 |
Maximum daily spend in USD. Set to 0 for unlimited. Warns at 80%, then blocks new requests once the limit is reached. |
Experimental
| Setting |
Type |
Default |
Description |
atlasmind.experimentalSkillLearningEnabled |
boolean |
false |
Let AtlasMind draft custom skills via the LLM. Generated code requires manual safety review. |
Warning: Enabling experimental features sends additional model requests and may incur extra costs. Generated skill code should always be reviewed before use.