memory-branches

name	memory-branches
description	Create and manage graph branches — isolated workspaces for staging changes before merging to main. Use for what-if analysis, planning, or any work that should not immediately affect the main graph.
metadata	{"author":"emergent","version":"1.0"}

Work with Memory graph branches — isolated copies of the graph where you can create, update, and delete objects without affecting the main branch until you explicitly merge.

Rules

• Project context is auto-discovered — the CLI walks up the directory tree to find .env.local containing MEMORY_PROJECT or MEMORY_PROJECT_ID. If .env.local is present anywhere above the current directory, --project is not needed. Only pass --project <id> explicitly when overriding or when no .env.local exists.
• Use --branch <id> on all graph write commands when working on a branch — objects and relationships created without --branch go to the main branch.
• Trust this skill over --help output — --help text may lag behind the installed binary.

---

What is a branch?

A branch is an isolated workspace scoped to a project. Objects and relationships created on a branch are invisible to the main branch (and vice versa) until you merge. Branches are useful for:

• What-if analysis — model a scenario without polluting the main graph
• Planning — stage a set of changes for review before committing
• Parallel work — multiple agents working on different aspects simultaneously

The main branch has no ID — it is the default when --branch is omitted.

---

Branch lifecycle

create branch → write objects/relationships with --branch <id> → verify → merge → delete branch

---

Step 1 — Create a branch

# Create a branch scoped to the current project:
memory graph branches create \
  --name "plan/add-auth-service" \
  --description "Staging area for auth service v2 design"

Output:

ID:       7602d370-64c2-451b-81a2-0b50ba74343a
Name:     plan/add-auth-service
Desc:     Staging area for auth service v2 design
Project:  ea62f9f7-396a-4b1e-912b-3b5579a7cf0a
Created:  2026-03-15T10:00:00Z

Capture the branch ID immediately:

BRANCH_ID=$(memory graph branches create \
  --name "plan/add-auth-service" \
  --description "Staging area for auth service v2 design" \
  --output json | python3 -c "import json,sys; print(json.load(sys.stdin)['id'])")
echo "Branch: $BRANCH_ID"

---

Step 2 — Write to the branch

Pass --branch <id> to every graph write command. Without it, writes go to the main branch.

# Create an object on the branch:
memory graph objects create \
  --type Service --key "svc-auth-v2" --name "auth-service-v2" \
  --status planned \
  --branch "$BRANCH_ID"

# Create a relationship on the branch:
memory graph relationships create \
  --type depends_on --from <src-entity-id> --to <dst-entity-id> \
  --branch "$BRANCH_ID"

# Batch-create objects on the branch (subgraph format):
# Note: --branch is not yet supported on create-batch; use single-create or
# set branch_id in the JSON body via --output json piped to curl if needed.

---

Step 3 — List branch contents

# List objects on the branch:
memory graph objects list --branch "$BRANCH_ID"

# List objects of a specific type on the branch:
memory graph objects list --type Service --branch "$BRANCH_ID"

# List relationships on the branch:
memory graph relationships list --branch "$BRANCH_ID"

# List all branches for the project:
memory graph branches list

---

Step 4 — Preview the merge (dry run)

Before merging, always preview what will change. The dry run classifies each diverged object:

Status	Meaning
added	Exists on source branch only — will be added to target
fast_forward	Changed on source only — will be updated on target
conflict	Changed on both branches — requires manual resolution
unchanged	Identical on both branches — no action taken

# Dry run (default — no changes made):
memory graph branches merge main \
  --source "$BRANCH_ID"

# Get full conflict details as JSON:
memory graph branches merge main \
  --source "$BRANCH_ID" --output json

Merging into main: Use the special keyword main as the target — the main graph has no branch ID and does not appear in branches list.

memory graph branches merge main --source "$BRANCH_ID"

---

Step 5 — Execute the merge

memory graph branches merge main \
  --source "$BRANCH_ID" --execute

Output:

Merge APPLIED
Source:  7602d370-64c2-451b-81a2-0b50ba74343a
Target:  <main-branch-id>

Objects (3 total):
  added:        2
  fast_forward: 1
  conflict:     0
  unchanged:    0

2 object(s) applied to target branch.

---

Step 6 — Delete the branch (cleanup)

After a successful merge, delete the branch to keep the list clean:

memory graph branches delete "$BRANCH_ID"

---

Full example — end to end

export MP=ea62f9f7-396a-4b1e-912b-3b5579a7cf0a

# 1. Create branch
BRANCH_ID=$(memory graph branches create \
  --name "plan/add-payment-service" --output json \
  | python3 -c "import json,sys; print(json.load(sys.stdin)['id'])")
echo "Working on branch: $BRANCH_ID"

# 2. Create objects on the branch
memory graph objects create \
  --type Service --key "svc-payments" --name "payments-service" \
  --status planned --branch "$BRANCH_ID"

memory graph objects create \
  --type ExternalDependency --key "dep-stripe" --name "Stripe" \
  --branch "$BRANCH_ID"

# 3. Wire relationships on the branch
PAYMENTS_ID=$(memory graph objects list \
  --type Service --branch "$BRANCH_ID" --output json \
  | python3 -c "import json,sys; d=json.load(sys.stdin); \
    print(next(o['entity_id'] for o in d['items'] if o.get('key')=='svc-payments'))")

STRIPE_ID=$(memory graph objects list \
  --type ExternalDependency --branch "$BRANCH_ID" --output json \
  | python3 -c "import json,sys; d=json.load(sys.stdin); \
    print(next(o['entity_id'] for o in d['items'] if o.get('key')=='dep-stripe'))")

memory graph relationships create \
  --type uses_dependency --from "$PAYMENTS_ID" --to "$STRIPE_ID" \
  --branch "$BRANCH_ID"

# 4. Verify branch contents
memory graph objects list --branch "$BRANCH_ID"

# 5. Preview merge into main (use the keyword "main" — no ID needed)
memory graph branches merge main \
  -- 


memory graph branches merge main \
  --  --execute


memory graph branches delete 

---

Branch management commands

# List all branches for the project (shows ID, name, description, parent, created):
memory graph branches list

# Get details for a specific branch:
memory graph branches get <branch-id>

# Rename a branch:
memory graph branches update <branch-id> --name "new-name"

# Update description only:
memory graph branches update <branch-id> --description "new purpose"

# Rename and update description:
memory graph branches update <branch-id> --name "new-name" --description "new purpose"

# Delete a branch:
memory graph branches delete <branch-id>

Fork a branch (copy-on-fork)

fork creates a new branch and immediately copies all HEAD objects and relationships from the source. This is different from create --parent, which only records lineage metadata without copying any data.

Copied objects preserve their canonical IDs so a subsequent merge back into the source is identity-aware.

# Fork from main — copies everything:
FORK_ID=$(memory graph branches fork main \
  --name "what-if/add-payment-service" --output json \
  | python3 -c "import json,sys; print(json.load(sys.stdin)['branch_id'])")

# Fork from an existing branch:
memory graph branches fork "$SOURCE_BRANCH_ID" \
  --name "child-branch" --description "subset of parent"

# Fork with type filter — only copy Service and API objects:
memory graph branches fork main \
  --name "services-only" \
  --filter-type Service \
  --filter-type API

Output:

Branch forked successfully.
Branch ID:             7602d370-64c2-451b-81a2-0b50ba74343a
Branch Name:           what-if/add-payment-service
Source:                main
Copied Objects:        42
Copied Relationships:  18

Flags:

• --name <name> — new branch name (required)
• --description <text> — optional description
• --filter-type <type> — only copy objects of this type (repeatable); relationships where one endpoint was excluded are skipped

When to use fork vs create:

	branches create	branches fork
Creates a branch	✓	✓
Records parent lineage	via --parent	✓ (automatic)
Copies objects/relationships	✗	✓
Use for	empty staging area	what-if with existing data

Querying a branch

Use memory query --mode=search --branch <id> to search a specific branch. Without --branch, the main graph is searched. --branch is not supported in agent mode.

# Search the main graph:
memory query --mode=search "planned services"

# Search a specific branch:
memory query --mode=search --branch "$BRANCH_ID" "planned services"

---

Merge edge cases

Situation	What happens
--execute with conflicts	Blocked — merge is not applied. Resolve conflicts manually then retry.
--execute with no conflicts	Runs in a single DB transaction. All-or-nothing: if any write fails, the whole merge rolls back.
Object added on branch, relationship points to it	The relationship's src_id/dst_id is remapped to the new canonical ID on the target branch automatically.
Same property key, same value on both branches	Not a conflict — classified as fast_forward or unchanged.
Same property key, different values on both branches	Conflict — listed under conflicts in --output json.
status, key, or labels changed on branch	Correctly detected and merged (included in content hash).
Graph > 500 objects	Preview is truncated — a warning is printed. Run with --output json to see all. Merge execution is not affected by the preview limit.
Merge of an already-merged branch	Objects that are identical on both sides are classified unchanged and skipped — safe to re-run.

Common mistakes

Mistake	Fix
Forgetting --branch on writes	Objects silently go to main branch — always pass --branch $BRANCH_ID
Using --branch on branches create	branches create doesn't take --branch; use --parent for child branches
Trying MEMORY_BRANCH env var	Not supported — always pass --branch <id> explicitly
Trying X-Branch-ID header	Not a header — branch is a body field (for create) or query param (for list)
Merging without a dry run first	Always preview with branches merge <target> --source <src> before --execute
Not capturing the branch ID	Run branches list to recover it, or always capture at creation time
Using fork output id field	fork returns branch_id, not id — use json.load(sys.stdin)['branch_id']

---

See also

• memory-graph — creating and updating objects and relationships
• memory-query — querying the graph with natural language