chore: initialize project scaffold and config

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

mnemosyne.toml

@@ -0,0 +1,78 @@
+# Mnemosyne configuration
+# Copy this to your working directory and customize.
+
+[proxy]
+host = "127.0.0.1"
+port = 0  # 0 = random free port
+upstream = "https://api.anthropic.com"
+
+[paging]
+window_size = 200000
+enable_paging = true
+enable_trim = true
+age_threshold = 4
+min_evict_size = 500
+
+[fidelity]
+# Pressure zone thresholds (as fraction of window_size)
+normal_max = 0.50
+caution_max = 0.70
+warning_max = 0.85
+critical_max = 0.95
+# Above critical_max = emergency zone
+
+# Fidelity degradation settings
+pin_duration_turns = 5  # How long a faulted object stays pinned
+min_age_for_degrade = 3  # Don't degrade objects from the last N turns
+
+[helper_llm]
+model = "claude-haiku-4-5-20251001"
+# If empty, uses ANTHROPIC_API_KEY env var
+api_key = ""
+base_url = "https://api.anthropic.com"
+timeout_seconds = 10
+max_retries = 2
+# Max tokens for summary generation
+max_summary_tokens = 1024
+max_compact_tokens = 256
+max_stub_tokens = 64
+max_micro_fault_tokens = 200
+max_goal_tokens = 128
+
+[object_store]
+# "memory" for in-memory only (Phase 1-2)
+# "sqlite" for local file (Phase 2+)
+# "postgresql" for full backing store (Phase 3+)
+backend = "memory"
+
+[object_store.sqlite]
+path = "mnemosyne.db"
+
+[object_store.postgresql]
+host = "localhost"
+port = 5433
+database = "mnemosyne"
+user = "mnemosyne"
+password = "mnemosyne_dev"
+
+[embeddings]
+model = "all-MiniLM-L6-v2"
+dimension = 384
+# Use ONNX runtime for faster inference
+use_onnx = true
+
+[admission]
+enabled = false  # Enable in Phase 4d
+threshold = 0.4
+weight_type = 0.35
+weight_novelty = 0.25
+weight_utility = 0.25
+weight_recency = 0.15
+
+[entropy]
+enabled = false  # Enable in Phase 4e
+normal_max = 1.5
+elevated_max = 2.2
+# Above elevated_max = high entropy
+window_size = 20  # Rolling window of tokens for entropy calculation
+debounce_count = 3  # Require N consecutive high-entropy tokens