Skip to main content

Construct tab

The Construct tab exposes every Engram runtime config field, grouped exactly like the UI. Each field has a description, use cases, safe default, validation, and risk warning.

Caution:

Fields marked risky can disable Engram, change save targets, change Git behavior, or affect memory security. Read the warning before changing them.

Core group

Enabled

Config key: enabled
Control: toggle
Default: true
Risk: risky

Master switch. Disabling it stops Engram behavior entirely. Use only for temporary shutdown or testing.

Save Target

Config key: scope
Control: select — workspace, global, both
Default: both
Risk: risky

Controls where new approved memories are saved. Use workspace for repo-specific memory, global for personal/team memory, both for fresh installs that want both.

Read Mode

Config key: read
Control: select — auto, startup, always, manual, off
Default: auto
Risk: normal

Controls when agent hooks inject memory context. auto loads on session start and reinjects only when routed context changes. manual and off reduce automation at the cost of context bloat.

Proof Mode

Config key: proof
Control: select — off, compact
Default: off
Risk: normal

Whether hooks append a compact Engram proof: line on each eligible turn. Useful for debugging and audit visibility.

Global Memory Path

Config key: global_path
Control: text/path
Default: empty until configured
Risk: risky

Filesystem path for global memory. Use a stable, user-owned folder such as ~/Documents/engram. Avoid temp folders, synced public folders, and directories you cannot write to.

Warning:

Using a cloud-synced public folder for private memory can leak secrets. Use a private path or a private Git repo.

CLI equivalent:

engram update-global-folder ~/Documents/engram
engram ugf ~/Documents/engram

Default Profile

Config key: default_profile
Control: select
Default: empty
Risk: risky

Profile used when none is explicitly set. See Profiles and scope resolution.

Active Roles

Config key: roles
Control: roles/comma input
Default: empty list
Risk: normal

Restricts and reranks memories by role. Use safe names matching ^[a-zA-Z0-9][a-zA-Z0-9._-]{0,63}$.

Load Routing group

Load Limit

Config key: load.limit
Control: number 1–32
Default: 8
Risk: normal

Max memories returned by normal load. Lower values reduce context bloat for low-context models; higher values help deep architecture tasks.

Memory Limits group

Rule Line Target

Config key: memory.rule_line_target
Control: number 50–200, step 10
Default: 70
Risk: normal

Recommended size for rule memories. Concise rules route better than overlong policies.

Rule Line Hard Limit

Config key: memory.rule_line_hard_limit
Control: number 50–200, step 10
Default: 100
Risk: risky

Hard maximum for rule memories.

Warning:

Raising this can increase context bloat and reduce routing quality. Keep rules concise.

Graph group

graph.enabled

Control: toggle
Default: true
Risk: normal

Enables dependency/relationship routing via depends_on, related memories, and the graph view.

Control: number 1–20
Default: 4
Risk: normal

Limits related memories pulled through graph signals.

Control: number 0–1, step 0.01
Default: 0.22
Risk: normal

Minimum similarity score for related edges. Raise for precision, lower for recall.

Vector Search group

vector.enabled

Control: toggle
Default: true
Risk: normal

Enables optional local vector routing. No cloud dependency.

vector.auto_threshold

Control: number 10–1000
Default: 100
Risk: normal

Memory count where vector search activates. Small vaults may not need vector search.

vector.candidate_pool

Control: number 8–100
Default: 24
Risk: normal

How many candidates vector search considers before reranking. Higher improves recall at latency cost.

vector.dimensions

Control: number 16–512
Default: 64
Risk: normal

Embedding dimensions for the local vector sidecar. Changing this requires a rebuild.

Rule Variants group

rule_variants.enabled

Control: toggle
Default: false
Risk: normal

Enables role/strictness variants. Use when teams need light/balanced/strict routing.

rule_variants.active

Control: select — light, balanced, strict
Default: balanced
Risk: normal

Controls strictness of loaded rules. strict helps lower-tier models; light/balanced usually suit stronger models.

Live Sync group

live_sync.enabled

Control: toggle
Default: false
Risk: normal

Syncs generated agent context files on save.

Global Git group

Warning:

All Global Git fields are risky. They control audit history and team sync behavior for global memory. Review each before enabling.

FieldControlDefaultNotes
global_git.enabledtoggletrueEnables Git behavior for global memory
global_git.remotetextoriginGit remote name; cannot contain whitespace
global_git.remote_urltextemptyShared global memory remote URL; HTTPS/SSH accepted
global_git.branchtextmainTarget branch for sync
global_git.auto_synctoggletrueAuto pull/push behavior
global_git.auto_resolvetoggletrueAuto conflict handling — review memory diffs

Pattern Mining group

FieldControlDefaultNotes
pattern_mining.enabledtogglefalseExperimental recurring-pattern extraction
pattern_mining.thresholdnumber 1–203Repetitions before a pattern candidate matters
pattern_mining.lookback_sessionsnumber 1–10020Recent sessions to inspect

PR Workflow group

FieldControlDefaultNotes
pr_workflow.enabledtogglefalseExperimental team PR workflow for memory changes
pr_workflow.target_branchtextmainBranch receiving memory PRs

Encryption group

Warning:

Encryption config exists, but encrypted storage is not implemented yet. Document current limitations clearly to users.

FieldControlDefaultNotes
encryption.enabledtogglefalseFuture/advanced encryption mode
encryption.scopeselect — workspace, globalglobalWhich scope encryption applies to
encryption.key_sourceselect — portable-fileportable-fileKey source strategy; backup loss risk

Next steps