Workflow Modes

Snipara offers two workflow modes optimized for different task complexities. Choose LITE for quick fixes, FULL for complex features that need durable memory, automation, and review.

Token Savings: LITE mode uses ~3-5K tokens. FULL mode uses ~8-15K tokens but adds session continuity, durable memory capture, and reviewable automation.

Quick Decision

Ask yourself these 4 questions:

Question	If Yes
Will this take multiple sessions?	+1 toward FULL
Does it affect 5+ files?	+1 toward FULL
Am I making architectural decisions?	+1 toward FULL
Will others need to understand this later?	+1 toward FULL

Score 0-1 = LITE mode | Score 2+ = FULL mode

LITE Mode (Default)

Use for bug fixes, small features, single-session work, and tasks where you know exactly which files to modify.

LITE MODE WORKFLOW
──────────────────

1. rlm_context_query("your task")     # Get relevant context
2. Read specific files                 # Direct file access
3. Make changes                        # Edit code
4. Run tests locally                   # pnpm test / pytest

Token budget: ~3-5K from Snipara

LITE Mode Example

rlm_context_query(query="fix null check authentication")

# Read the specific file

# Edit to add null check

pnpm test

FULL Mode (Complex Features)

Use for multi-day features, architectural changes, team coordination, and documentation-heavy work that spans multiple sessions.

FULL MODE WORKFLOW
──────────────────

PHASE 1: CONTEXT GATHERING
├── rlm_shared_context()              # Team standards
├── rlm_recall("feature area")        # Past decisions
└── rlm_context_query(max_tokens=8000)# Deep context

PHASE 2: PLANNING
├── rlm_plan("feature description")   # Execution plan
├── rlm_decompose("feature")          # Break into tasks
└── rlm_remember(type="decision")     # Store key choices

PHASE 3: EXECUTION LOOP
├── rlm_inject("current context")     # Session focus
├── rlm_multi_query([...])            # Batch queries
├── execute_python(code)              # Test logic
└── Edit files locally                # Actual changes

PHASE 4: PERSIST DURABLE OUTCOMES
├── rlm_remember_if_novel(...)        # Skip duplicates
├── rlm_end_of_task_commit(...)       # Capture durable task outcome
└── Review inbox in dashboard         # When project policy = INBOX

Token budget: ~8-15K from Snipara

FULL Mode Example

# Phase 1: Context

rlm_shared_context(categories=["BEST_PRACTICES"])

rlm_recall(query="rate limiting decisions")

rlm_context_query(query="API middleware", max_tokens=8000)

# Phase 2: Plan

rlm_plan(query="implement rate limiting for API")

rlm_remember(type="decision", content="Using Redis sliding window")

# Phase 4: Implement

rlm_inject(context="Working on rate limiting, Redis backend")

rlm_multi_query(queries=[{query: "Redis patterns"}, ...])

# Phase 5: Persist durable outcome

rlm_end_of_task_commit(summary="Implemented Redis sliding-window rate limiting and left tests as the next step")

Reviewable automation: In FULL mode, projects can route automated memory writes into an inbox instead of recalling them immediately. This keeps compaction and commit hooks useful without polluting future sessions.

Mode Selection Examples

Task	Mode	Why
Fix typo in README	LITE	Single file, obvious change
Fix null check in auth.ts	LITE	Known file, small fix
Add loading spinner	LITE	Single component
Add rate limiting to API	FULL	Multi-file, architectural
Refactor auth to JWT	FULL	Breaking change, multi-file
New billing integration	FULL	New feature, external API
Multi-tenant support	FULL	Architectural, multi-session

Session Continuity (FULL Mode)

FULL mode enables work that spans multiple sessions through the memory system.

Starting a New Session

rlm_recall(query="feature-name progress status")

rlm_context_query(query="feature-name")

Ending a Session

rlm_end_of_task_commit(
  summary="Feature X: Completed steps 1-3. Next: implement Y. Blocker: Z",
  outcome="partial",
  persist_types=["decision", "learning", "workflow"]
)

Tools Reference

Need	Tool	Mode
Quick answer	`rlm_ask`	Both
Deep context	`rlm_context_query`	Both
Past decisions	`rlm_recall`	FULL
Team standards	`rlm_shared_context`	FULL
Plan complex work	`rlm_plan`	FULL
Break down task	`rlm_decompose`	FULL
Save decision	`rlm_remember`	FULL
Save only novel memory	`rlm_remember_if_novel`	FULL
Persist durable task outcome	`rlm_end_of_task_commit`	FULL
Batch queries	`rlm_multi_query`	FULL
Test logic	`execute_python`	FULL

MCP Tools Reference Learn About Memory