Practical overview
When it fits
Use ICM when coding sessions or tools need access to the same persistent project context, or when you want to add and retrieve memories directly from a CLI.
Who it is for
- Developers who use multiple AI coding assistants and want shared project context.
- Teams or individuals who want to store and recall coding decisions, fixes and conventions through a CLI or MCP-capable tools.
Documented uses
- Store a project fact with a topic and importance.
- Recall relevant project context using a question, topic or keyword.
- Add memory instructions, skills and hooks to detected agent configurations.
- Extract facts from coding-agent tool output into queued memories.
Requirements & limits
The repository is public and licensed under Apache-2.0. ICM is provided as a command-line tool; its README documents installation options for macOS, Windows, Linux and Nix.
- Standard initialization configures instructions, skills and hooks;
--mode alladditionally registers the MCP server. - Keyword recall avoids downloading an embedding model; semantic search downloads one on first use.
- Automatic extraction from tool output is separate from storage and recall: it can use an installed agent CLI or be configured with
provider = "none".
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Initialize ICM for detected coding agents
Install ICM's instructions, skills and hooks for agents it detects, then optionally register its MCP server.
Before you start
- ICM installed as a command-line tool.
- A supported coding-agent tool detected by ICM.
Steps
Run
icm initto add instructions, skills and hooks for detected agents.To add MCP registration as well, run
icm init --mode all.
Expected outcome
Detected agents receive ICM configuration; `--mode all` also registers the MCP server in tools that support it.
Guide limitations
- The standard mode does not register an MCP server.
- Aider requires
--per-projectfor its per-project conventions file when using the documented all-tools setup.
Store and recall a project memory
Add a project fact to ICM and retrieve it with a query.
Before you start
- ICM installed and initialized.
Steps
Run
icm storewith a topic, content and importance for the fact.Run
icm recallwith a question or search phrase about that fact.Use keyword recall with
--no-embeddingsif you do not want to download the semantic-search model.
Expected outcome
The CLI stores the fact in ICM and returns matching memory details for the recall query.
Guide limitations
- The first semantic-search use downloads a multilingual embedding model of about 2 GB.
- Keyword recall works without that model but does not use semantic embeddings.
Compare options for AI coding assistance
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Icm | Store a project fact with a topic and importance. | CLI, Self-hosted, Server | The repository is public and licensed under Apache-2.0. ICM is provided as a command-line tool; its README documents installation options for macOS, Windows, Linux and Nix. |
| Continuum — pouyahasanamreji | Store project lessons so coding agents can retrieve them across conversations. | CLI, Self-hosted, Server | The repository documents local Docker Compose and bare-metal development setups. Bare-metal startup requires Node 22.12.0 or later and pnpm 10.33 or later; the Compose path skips these host prerequisites. |
| Haft | Record the problem a project is solving and the options considered. | CLI, Self-hosted |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- rtk-ai/icm · Official homepage title · retained in its source language