SedimentPalace MCP is a local-first memory engine for AI agents with layered sedimentation, policy controls, journaling, recovery, and telemetry.
- Layered memory model (
01_Shallow,02_Sediment,03_Bedrock) - Structured MCP tool responses (
result.data/result.error) - Policy engine (confirm checks, rate limits, budgets)
- Operation journal + recovery flow
- Per-tool telemetry and metrics snapshot
- Healthcheck endpoint for runtime readiness
python -m pip install -e ".[dev]"
pytest -q
python -m sediment_palace.mainInitialize a target workspace with local agent configs, memory folders, and instruction files:
start.bat --workspace D:\Projects\MyWork --agents codex,claude,qwen,opencode,kimiWhat this creates in the workspace:
memory/layout (01_Shallow,02_Sediment,03_Bedrock,_System,_Archive,PALACE_MAP.md)AGENT_BOOTSTRAP.md- Agent instruction files (for selected agents):
AGENTS.md,CODEX.md,CLAUDE.md,QWEN.md,OPENCODE.md,KIMI.md - Local agent config files (
.codex/config.toml,.qwen/settings.json,.opencode/opencode.json,.kimi/settings.json,.mcp.json)
Global config mode is optional:
start.bat --scope global --workspace D:\Projects\MyWork --agents codex,claudeSet via environment variables:
SEDIMENT_PROJECT_ROOT(default:.)SEDIMENT_TIMEOUT_DEFAULT_SECONDS(default:2.0)SEDIMENT_TIMEOUT_METABOLIZE_SECONDS(default:5.0)SEDIMENT_BUDGET_WRITE_CONTENT(default:50000)SEDIMENT_BUDGET_READ_QUERY(default:512)SEDIMENT_BUDGET_SEARCH_QUERY(default:512)SEDIMENT_BUDGET_SEARCH_ROOM(default:1024)SEDIMENT_BUDGET_PURGE_REASON(default:1024)
- Language index:
docs/README.md - English docs:
docs/en/README.md - Russian docs:
docs/ru/README.md - Plug-and-play bootstrap prompt:
AGENT_BOOTSTRAP.md
python -m ruff check src testspython -m mypy srcpytest -q
CI runs all gates on push and pull requests.
LICENSECONTRIBUTING.mdCODE_OF_CONDUCT.mdSECURITY.mdSUPPORT.md.github/ISSUE_TEMPLATE/*.github/PULL_REQUEST_TEMPLATE.md
- Latest RC tag:
v0.1.0-rc1 - For release process, use Git tags and GitHub Releases.