Skip to main content
MIRA ships with two reasoning engines. The active engine is shown in the status bar at the bottom of the window. Switch between them in Settings → Engine (⌘,).

​
Engine comparison

DimensionNative Agent Engine (NAE)RLM Engine
Default✅ YesNo
LLM accessDirect API callsVia Python bridge
Context managementBuilt-in — token budget, compaction, summarisation, episodic memoryDelegated to RLM layer
Code executionNoYes — local Python REPL (full OS access)
Multi-agentSingle / Multi / Auto modesNot applicable
ProvidersOpenAI, Anthropic, Ollama, BedrockBedrock, Anthropic, OpenAI, Ollama
Best forOpen-ended research, long sessions, multi-domain tasksData analysis, calculations, tasks where provably correct answers matter
StreamingToken-by-token chunksIteration-by-iteration updates
Skills✅ Supported✅ Supported
Workflows✅ Supported✅ Supported
MCP tools✅ Supported✅ Supported
Document context✅ Supported✅ Supported
Switching engines resets the active session — context from one engine is not transferred to the other.

​
Engine status

The active reasoning engine is always visible in the status bar at the bottom of the MIRA window.
StatusMeaning
● RunningEngine is ready and accepting queries
◐ InitializingEngine is starting up (normal on first launch, ~60 s)
↻ RestartingEngine crashed and is being restarted with exponential backoff
✕ FailedEngine could not start — check Settings → Engine and your credentials
Switch engines any time in Settings → Engine (⌘,). Switching resets the active session.

​
Switching engines

Switching engines resets the active session — context, history, and document context from one engine are not carried over to the other. This is intentional: the two engines maintain fundamentally different internal state. To switch:
  1. Open Settings (⌘,)
  2. Go to the Engine tab
  3. Click the engine you want to activate
  4. A confirmation dialog appears if there is an active session
  5. The status bar updates immediately
Your engine choice is persisted across restarts.

​
Configuration

Each engine has its own configuration panel in Settings → Engine:

​
Deep dives

Native Agent Engine

How NAE manages token budgets, compacts context, spawns sub-agents, and maintains episodic memory

RLM Engine

How RLM writes code, executes it, interprets output, and iterates to a verified answer