Skip to content

Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors

Rust

Merge branch 'main' of https://github.com/akitaonrails/ai-memory into feat/kimi-code-support

Lucas Oliveiraa817fcf09ea7
diff --cc AGENTS.md
index 5dbe8081,7fa8fb5a..9c55d453
--- a/AGENTS.md
+++ b/AGENTS.md
@@@ -36,8 -36,8 +36,8 @@@ install or refresh work
If you're about to write a durable project rule ("always X", "never
Y", "all PRs must ..."), write it in the project's canonical agent instruction file.
Many projects use CLAUDE.md for Claude Code and
- AGENTS.md for Codex / OpenCode / Cursor / Gemini CLI / Kimi Code,
-AGENTS.md for Codex / OpenCode / Cursor / Gemini CLI / Grok Build CLI, but if the project
-says one file is canonical, use that file.
++AGENTS.md for Codex / OpenCode / Cursor / Gemini CLI / Grok Build CLI / Kimi Code,
+but if the project says one file is canonical, use that file.
If the rule is a standing *user/team* preference that should apply to
every project (tech choices, code style, personal conventions), save it
@@@ -53,8 -53,8 +53,9 @@@ latest binary's recommended copy
- **From the agent** (no terminal needed): ask "refresh the ai-memory
routing in this project". The agent calls `memory_install_self_routing`,
picks the right filename for itself (Claude Code -> `CLAUDE.md`; Codex /
- OpenCode / Cursor / Gemini / Kimi Code -> `AGENTS.md`), uses its Write /
- Edit tool to replace or append the returned `markered_block` while preserving
- OpenCode / Cursor / Gemini / Grok -> `AGENTS.md`), uses its Write / Edit tool
- to replace or append the returned `markered_block` while preserving
++ OpenCode / Cursor / Gemini / Grok -> `AGENTS.md`; Kimi Code -> `AGENTS.md`),
++ uses its Write / Edit tool to replace or append the returned
++ `markered_block` while preserving
non-ai-memory user content, then writes or updates each returned
`managed_skills` item under the selected skill root from `target_hints`
using its `relative_path`.
diff --cc CHANGELOG.md
index f58b0bd2,fa435077..ea432a7a
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@@ -8,20 -8,31 +8,43 @@@ and this project adheres to [Semantic V
## [Unreleased]
### Added
+- Kimi Code CLI is now a supported MCP + lifecycle-hook integration
+ (`--client kimi-code` / `--agent kimi-code`, alias `kimi`). `install-mcp`
+ merges an `mcpServers` entry into `~/.kimi-code/mcp.json` with a plain `url`
+ (Kimi Code treats `url` with no `transport` field as streamable HTTP) plus
+ optional bearer `headers`. `install-hooks` merges `[[hooks]]` entries into
+ `~/.kimi-code/config.toml`, preserving the provider/model settings the same
+ file holds, and covers 9 events: `SessionStart`, `SessionEnd`,
+ `UserPromptSubmit`, `PreToolUse`, `PostToolUse`, `Stop`, `SubagentStart`,
+ `SubagentStop`, and `PreCompact`. Both paths honor `$KIMI_CODE_HOME`.
+ Handoff injection happens at `SessionStart` through hook stdout, which Kimi
+ Code appends to the model context; `setup-agent` and `uninstall` handle the
+ new agent like the other first-class integrations.
+ - Grok Build CLI is now a first-class MCP client as well as a hook agent:
+ `install-mcp --client grok` renders and `--apply` merges a native HTTP
+ entry into `$GROK_HOME/config.toml` (default `~/.grok/config.toml`)
+ (`[mcp_servers.ai-memory]` with
+ `url` / `enabled` / `[mcp_servers.ai-memory.headers]` — not Codex's
+ `http_headers` key). `install-hooks --agent grok` reuses that entry to
+ infer server URL and bearer token; `uninstall` strips the MCP table;
+ `setup-agent --agent grok` prints the companion `install-mcp` tip.
+ Managed routing skills can target `.grok/skills` / `$GROK_HOME/skills`
+ (default `~/.grok/skills`) via
+ `install-skills --agent grok` and `memory_install_self_routing`
+ `target_hints`. Lifecycle capture was already present; handoff injection
+ remains unavailable because Grok ignores SessionStart stdout — recover
+ via MCP `memory_handoff_accept`.
### Changed
+ - `install-hooks --agent devin` now infers the server URL and bearer token from
+ `~/.devin/config.json` when flags and environment settings are absent, keeping
+ hooks aligned with an existing Devin MCP registration.
+ - Grok routing guidance now treats Grok Build CLI as an AGENTS-based client and
+ directs its managed skills to `.grok/skills` or `$GROK_HOME/skills` (default
+ `~/.grok/skills`). The MCP self-routing payload and installed prompt surface
+ now guard those targets;
+ Grok (like Zero) still resumes handoffs through `memory_handoff_accept`
+ because it ignores SessionStart stdout.
- One-shot client commands (`rename-project`, `status`, `write-page`, …) no
longer print the "cannot write log files … falling back" warning when the
log directory is unwritable — they still degrade to temp/stderr logging
diff --cc README.md
index ee9c84d6,18f1b218..10af4974
--- a/README.md
+++ b/README.md
@@@ -33,9 -33,8 +33,9 @@@
| Claude Desktop | MCP-only | Uses `mcp-remote`; no lifecycle hooks. |
| OpenClaw | Supported | MCP config + native plugin lifecycle hooks. |
| Antigravity CLI | Supported | MCP config (`serverUrl`) + lifecycle hooks (`agy` alias). |
- | Grok Build CLI | Hooks | Lifecycle hooks via `install-hooks --agent grok` (`~/.grok/hooks/ai-memory.json`, Grok-specific hook bundle, native `--agent grok`). Capture works; no handoff injection — Grok ignores `SessionStart` stdout, so recover handoffs via MCP `memory_handoff_accept`. |
+ | Grok Build CLI | Supported | MCP config (`install-mcp --client grok` → `$GROK_HOME/config.toml`, default `~/.grok/config.toml`) + lifecycle hooks (`install-hooks --agent grok` → `$GROK_HOME/hooks/ai-memory.json`, default `~/.grok/hooks/ai-memory.json`, Grok-specific hook bundle). Capture works; no handoff injection — Grok ignores `SessionStart` stdout, so recover handoffs via MCP `memory_handoff_accept`. Skills root: `.grok/skills` / `$GROK_HOME/skills` (default `~/.grok/skills`). |
| Zero | Supported | `install-mcp --client zero` (native HTTP + bearer in `~/.config/zero/config.json`) + lifecycle hooks via `install-hooks --agent zero --apply` (exec-form native commands in `~/.config/zero/hooks.json`, JSON payload on stdin, no shell). Capture works incl. specialist (subagent) events; no handoff injection — Zero discards `sessionStart` stdout, so recover handoffs via MCP `memory_handoff_accept`. |
+| Kimi Code | Supported | MCP config (`url` entry in `~/.kimi-code/mcp.json`) + lifecycle hooks (`[[hooks]]` in `~/.kimi-code/config.toml`, 9 events including subagent start/stop); both paths honor `$KIMI_CODE_HOME`. Handoffs inject via `SessionStart` stdout. |
| VS Code Copilot | MCP-only | `.vscode/mcp.json` for Copilot agent mode; no lifecycle hooks (Copilot does not expose them yet). |
| Hermes Agent | Community | A community-maintained [`ai-memory-hermes-plugin`](https://github.com/MrLuciano/ai-memory-hermes-plugin) is available. It is not part of ai-memory's first-party install surface; review its compatibility matrix, install/uninstall scripts, and secret handling before using it. |
| LLM/auth providers | Supported | Anthropic, OpenAI, OpenAI OAuth/Codex, GitHub Copilot, Gemini, OpenCode Zen/Go, OpenAI-compatible endpoints, and generic OIDC device auth for native hooks. |
@@@ -271,8 -271,8 +272,8 @@@ docker run -d --name ai-memory
# 3. Wire your agent CLI in two commands. The wrapper takes care of
# mounts + auto-detecting ~/.claude/settings.json. Re-run with
# `--agent codex`, `--agent devin`, `--agent opencode`, `--agent gemini-cli`,
- # `--agent kimi-code`, `--agent omp`, `--agent oh-my-pi`, `--client cursor`,
- # `--client gemini-cli`, etc.
-# `--agent grok`, `--agent omp`, `--agent oh-my-pi`, `--client cursor`,
++# `--agent grok`, `--agent kimi-code`, `--agent omp`, `--agent oh-my-pi`, `--client cursor`,
+ # `--client gemini-cli`, `--client grok`, etc.
# for additional agents; full list in docs/install.md.
ai-memory install-mcp --client claude-code --apply
ai-memory install-hooks --agent claude-code --apply
diff --cc crates/ai-memory-cli/src/cli.rs
index 8429d472,4a60faa8..6bdaf9f2
--- a/crates/ai-memory-cli/src/cli.rs
+++ b/crates/ai-memory-cli/src/cli.rs
@@@ -937,9 -936,12 +940,15 @@@ pub enum McpClient
Zero,
/// Devin CLI — `~/.devin/config.json`.
Devin,
+ /// xAI Grok Build CLI — `~/.grok/config.toml` under
+ /// `[mcp_servers.<name>]` with native HTTP `url` + `headers`.
+ /// Pair with `install-hooks --agent grok` for lifecycle capture.
+ /// Grok ignores SessionStart stdout, so handoffs are recovered via
+ /// MCP `memory_handoff_accept` rather than hook injection.
+ Grok,
+ /// Kimi Code CLI (Moonshot AI).
+ #[value(alias = "kimi")]
+ KimiCode,
/// VS Code GitHub Copilot (agent mode) — per-workspace
/// `.vscode/mcp.json`. Copilot's agent mode reads MCP servers
/// from VS Code's own MCP framework (top-level `servers` key),
diff --cc crates/ai-memory-cli/src/commands/install_hooks.rs
index 9244a48e,a89043a9..8cbe2004
--- a/crates/ai-memory-cli/src/commands/install_hooks.rs
+++ b/crates/ai-memory-cli/src/commands/install_hooks.rs
@@@ -391,32 -358,77 +389,82 @@@ fn apply_base_path_to_hook_url(url: &st
}
}
- fn infer_installed_mcp_config(agent: AgentChoice) -> Option<InferredMcpConfig> {
- let client = mcp_client_for_agent(agent)?;
- let path = install_mcp::mcp_config_path(client).ok()?;
- let content = fs::read_to_string(path).ok()?;
- match client {
- McpClient::ClaudeCode => {
- infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "url")
- }
- McpClient::Codex => infer_codex_mcp_config(&content),
- McpClient::OpenCode => infer_json_mcp_config(&content, &["mcp", "ai-memory"], "url"),
- McpClient::Cursor => infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "url"),
- McpClient::GeminiCli => {
- infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "httpUrl")
- }
- McpClient::Openclaw => {
- infer_json_mcp_config(&content, &["mcp", "servers", "ai-memory"], "url")
- }
- McpClient::Omp => infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "url"),
- McpClient::Zero => infer_json_mcp_config(&content, &["mcp", "servers", "ai-memory"], "url"),
- McpClient::Pi => None,
- McpClient::AntigravityCli => {
- infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "serverUrl")
+ fn infer_installed_mcp_config(agent: AgentChoice) -> Result<Option<InferredMcpConfig>> {
+ if agent == AgentChoice::Grok {
+ let cwd = std::env::current_dir()
+ .context("could not resolve current dir for Grok project configuration")?;
+ let repo_root = ai_memory_consolidate::discover_repo_root(&cwd).ok();
+ if let Some(project_config) = find_grok_project_overlay(&cwd, repo_root.as_deref()) {
+ anyhow::bail!(
+ "Grok project configuration {} may override GROK_HOME; pass explicit --server-url and --auth-token rather than inferring MCP settings",
+ project_config.display()
+ );
}
- McpClient::Devin => infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "url"),
- McpClient::KimiCode => infer_json_mcp_config(&content, &["mcpServers", "ai-memory"], "url"),
- McpClient::ClaudeDesktop => None,
+ }
+ let Some(client) = mcp_client_for_agent(agent) else {
+ return Ok(None);
+ };
+ let path = install_mcp::mcp_config_path(client)?;
+ let Ok(content) = fs::read_to_string(path) else {
+ return Ok(None);
+ };
+ match client {
+ McpClient::ClaudeCode => Ok(infer_json_mcp_config(
+ &content,
+ &["mcpServers", "ai-memory"],
+ "url",
+ )),
+ // Codex uses `http_headers`; Grok uses `headers`. The shared TOML
+ // inferencer accepts both.
+ McpClient::Codex => Ok(infer_toml_mcp_config(&content)),
+ McpClient::Grok => infer_grok_mcp_config(&content),
+ McpClient::OpenCode => Ok(infer_json_mcp_config(
+ &content,
+ &["mcp", "ai-memory"],
+ "url",
+ )),
+ McpClient::Cursor => Ok(infer_json_mcp_config(
+ &content,
+ &["mcpServers", "ai-memory"],
+ "url",
+ )),
+ McpClient::GeminiCli => Ok(infer_json_mcp_config(
+ &content,
+ &["mcpServers", "ai-memory"],
+ "httpUrl",
+ )),
+ McpClient::Openclaw => Ok(infer_json_mcp_config(
+ &content,
+ &["mcp", "servers", "ai-memory"],
+ "url",
+ )),
+ McpClient::Omp => Ok(infer_json_mcp_config(
+ &content,
+ &["mcpServers", "ai-memory"],
+ "url",
+ )),
+ McpClient::Zero => Ok(infer_json_mcp_config(
+ &content,
+ &["mcp", "servers", "ai-memory"],
+ "url",
+ )),
+ McpClient::Pi => Ok(None),
+ McpClient::AntigravityCli => Ok(infer_json_mcp_config(
+ &content,
+ &["mcpServers", "ai-memory"],
+ "serverUrl",
+ )),
+ McpClient::Devin => Ok(infer_json_mcp_config(
+ &content,
+ &["mcpServers", "ai-memory"],
+ "url",
+ )),
++ McpClient::KimiCode => Ok(infer_json_mcp_config(
++ &content,
++ &["mcpServers", "ai-memory"],
++ "url",
++ )),
+ McpClient::ClaudeDesktop => Ok(None),
// MCP-only client: no AgentChoice counterpart routes here.
// Reachable only if a future install_hooks flow targets VS
// Code Copilot directly.
@@@ -437,10 -482,11 +518,12 @@@ fn mcp_client_for_agent(agent: AgentCho
AgentChoice::Openclaw => Some(McpClient::Openclaw),
AgentChoice::AntigravityCli => Some(McpClient::AntigravityCli),
AgentChoice::Zero => Some(McpClient::Zero),
+ AgentChoice::Grok => Some(McpClient::Grok),
+ AgentChoice::Devin => Some(McpClient::Devin),
+ AgentChoice::KimiCode => Some(McpClient::KimiCode),
- // Grok and Devin manage their own MCP config under ~/.grok/ and ~/.devin/;
- // we don't auto-infer a hook server URL from it.
- AgentChoice::Pi | AgentChoice::Grok | AgentChoice::Devin => None,
+ // Pi bridges MCP through its generated extension, not a native
+ // mcp.json the installer can scrape.
+ AgentChoice::Pi => None,
}
}
diff --cc crates/ai-memory-cli/src/commands/install_mcp.rs
index b9ca4813,f6a8112e..e2ae4fdf
--- a/crates/ai-memory-cli/src/commands/install_mcp.rs
+++ b/crates/ai-memory-cli/src/commands/install_mcp.rs
@@@ -182,18 -181,17 +186,29 @@@ pub(crate) fn mcp_config_path(client: c
})
}
+/// Kimi Code's data dir: `$KIMI_CODE_HOME` when set (non-empty), else
+/// `~/.kimi-code`. The env value comes in as a parameter so tests can
+/// exercise both branches without mutating process env.
+fn kimi_code_home(env_override: Option<std::ffi::OsString>) -> Result<PathBuf> {
+ if let Some(dir) = env_override.filter(|value| !value.is_empty()) {
+ return Ok(PathBuf::from(dir));
+ }
+ Ok(home_dir()
+ .context("could not locate $HOME for config-file auto-detect")?
+ .join(".kimi-code"))
+}
+
+ /// Resolve Grok Build CLI's user configuration root. Grok honours
+ /// `GROK_HOME`; otherwise it uses `~/.grok`.
+ pub(crate) fn grok_home() -> Result<PathBuf> {
+ if let Some(path) = std::env::var_os("GROK_HOME").filter(|path| !path.is_empty()) {
+ return Ok(PathBuf::from(path));
+ }
+ Ok(home_dir()
+ .context("could not locate $HOME for Grok configuration")?
+ .join(".grok"))
+ }
+
/// Resolve the user-config file for this client. Honours
/// `--config-file` when provided, else uses the canonical default
/// per client.
diff --cc crates/ai-memory-cli/src/commands/setup_agent.rs
index 32f22137,2b21231a..312af3c2
--- a/crates/ai-memory-cli/src/commands/setup_agent.rs
+++ b/crates/ai-memory-cli/src/commands/setup_agent.rs
@@@ -35,10 -35,11 +35,11 @@@ use std::path::{Path, PathBuf}
use anyhow::{Context, Result, bail};
use crate::cli::{AgentChoice, SetupAgentArgs};
+ use crate::commands::install_mcp;
use crate::commands::render_shared::{
- ANTIGRAVITY_LIFECYCLE_EVENTS, ANTIGRAVITY_TOOL_EVENTS, CODEX_PROFILE, CURSOR_PROFILE,
- GEMINI_PROFILE, build_claude_code_payload, build_devin_payload, build_grok_payload,
- hook_script_for_current_platform,
+ ANTIGRAVITY_LIFECYCLE_EVENTS, ANTIGRAVITY_TOOL_EVENTS, CLAUDE_CODE_EVENTS, CODEX_PROFILE,
+ CURSOR_PROFILE, GEMINI_PROFILE, build_claude_code_payload, build_devin_payload,
+ build_grok_payload, hook_script_for_current_platform,
};
use crate::config::{Config, DEFAULT_SERVER_URL};
diff --cc crates/ai-memory-core/src/routing_snippet.rs
index 56c73008,9acaf42c..0c02fe62
--- a/crates/ai-memory-core/src/routing_snippet.rs
+++ b/crates/ai-memory-core/src/routing_snippet.rs
@@@ -64,8 -64,8 +64,8 @@@ install or refresh work
If you're about to write a durable project rule ("always X", "never
Y", "all PRs must ..."), write it in the project's canonical agent instruction file.
Many projects use CLAUDE.md for Claude Code and
- AGENTS.md for Codex / OpenCode / Cursor / Gemini CLI / Kimi Code,
-AGENTS.md for Codex / OpenCode / Cursor / Gemini CLI / Grok Build CLI, but if the project
-says one file is canonical, use that file.
++AGENTS.md for Codex / OpenCode / Cursor / Gemini CLI / Grok Build CLI / Kimi Code,
+but if the project says one file is canonical, use that file.
If the rule is a standing *user/team* preference that should apply to
every project (tech choices, code style, personal conventions), save it
@@@ -81,8 -81,8 +81,9 @@@ latest binary's recommended copy
- **From the agent** (no terminal needed): ask "refresh the ai-memory
routing in this project". The agent calls `memory_install_self_routing`,
picks the right filename for itself (Claude Code -> `CLAUDE.md`; Codex /
- OpenCode / Cursor / Gemini / Kimi Code -> `AGENTS.md`), uses its Write /
- Edit tool to replace or append the returned `markered_block` while preserving
- OpenCode / Cursor / Gemini / Grok -> `AGENTS.md`), uses its Write / Edit tool
- to replace or append the returned `markered_block` while preserving
++ OpenCode / Cursor / Gemini / Grok -> `AGENTS.md`; Kimi Code -> `AGENTS.md`),
++ uses its Write / Edit tool to replace or append the returned
++ `markered_block` while preserving
non-ai-memory user content, then writes or updates each returned
`managed_skills` item under the selected skill root from `target_hints`
using its `relative_path`.
diff --cc crates/ai-memory-mcp/src/server.rs
index 5cb0bb30,d45d0a12..c618c81c
--- a/crates/ai-memory-mcp/src/server.rs
+++ b/crates/ai-memory-mcp/src/server.rs
@@@ -2453,7 -2453,7 +2453,8 @@@ impl AiMemoryServer
"antigravity_cli": "AGENTS.md",
"zero": "AGENTS.md",
"devin": "AGENTS.md",
+ "kimi_code": "AGENTS.md",
+ "grok": "AGENTS.md",
"default": "AGENTS.md"
},
"managed_skills": managed_skills,
diff --cc docs/install.md
index 3eb036d1,8398ffce..341a9f12
--- a/docs/install.md
+++ b/docs/install.md
@@@ -563,38 -563,11 +563,39 @@@ differences
The `SessionStart` hook injects pending handoffs through Devin's
`hookSpecificOutput.additionalContext`. Real Devin `SessionStart` and
- `PostToolUse` payloads may omit `session_id` and `cwd`; ai-memory derives stable
- session/cwd context from its hook state and process environment so those events
- are still captured.
+ `PostToolUse` payloads may omit `session_id` and `cwd`; ai-memory now infers cwd
+ from `DEVIN_PROJECT_DIR` or the hook process working directory when the payload
+ omits it, and mints/reuses a per-host session id from hook state when necessary,
+ so those events are still captured. A payload-provided value always wins.
+### Kimi Code
+
+Kimi Code keeps MCP servers in `~/.kimi-code/mcp.json` and lifecycle hooks in
+`~/.kimi-code/config.toml`; both move together when `$KIMI_CODE_HOME` is set.
+The CLI also accepts `--agent kimi` as an alias.
+
+```bash
+ai-memory install-mcp --client kimi-code --apply \
+ --server-url "http://homelab:49374/mcp" \
+ --auth-token "$TOKEN"
+
+ai-memory install-hooks --agent kimi-code --apply \
+ --server-url "http://homelab:49374" \
+ --auth-token "$TOKEN"
+```
+
+`install-hooks` merges `[[hooks]]` entries into `config.toml`, preserving the
+provider/model settings the same file holds, and stages the hook scripts under
+`~/.local/share/ai-memory/hooks/kimi-code/`. Capture covers 9 events —
+`SessionStart`, `SessionEnd`, `UserPromptSubmit`, `PreToolUse`, `PostToolUse`,
+`Stop`, `SubagentStart`, `SubagentStop`, and `PreCompact` — as fire-and-forget
+POSTs to `/hook`. A pending handoff is injected at `SessionStart` through the
+hook's stdout, which Kimi Code appends to the model context.
+
+Kimi Code hook entries accept only `event`, `matcher`, `command`, and
+`timeout`; extra fields make the whole `config.toml` fail to load, so prefer
+`install-hooks --apply` over hand edits.
+
### OpenCode
```bash
diff --cc docs/usage.md
index aedb4cd2,4f070322..675cb153
--- a/docs/usage.md
+++ b/docs/usage.md
@@@ -290,7 -292,7 +295,7 @@@ docker exec ai-memory git -C /data/wik
Durable project rules belong in the agent's rules file, not only in the
wiki. For Claude Code that is `CLAUDE.md`; for Codex, Devin CLI, OpenCode,
- Cursor, Kimi Code, and Gemini CLI it is usually `AGENTS.md`.
-Cursor, Gemini CLI, and Grok Build CLI it is usually `AGENTS.md`.
++Cursor, Gemini CLI, Grok Build CLI, and Kimi Code it is usually `AGENTS.md`.
The consolidator classifies compiled observations as `decision`,
`fact`, `rule`, or `gotcha`. Rule-tagged pages are routed to