agents-remember

Providers

Agents Remember ships two optional context providers. They are opt-in: enable them in the MCP settings, and they run as Docker services.

Provider id Capability What it indexes
codegraphcontext-code (CGC) code-relationship search your code repository
grepai-memory (grepai) semantic-memory search the memory / onboarding

CGC answers structural questions (who calls this, what does this depend on); grepai answers “where is the memory about X” over onboarding content.

Requirements

If Docker is not running, provider install and start will fail; if Ollama is not reachable, grepai will report degraded and index nothing.

Enable

Add the providers you want to the MCP settings providers block:

{
  "providers": {
    "codegraphcontext-code": {},
    "grepai-memory": {}
  }
}

Omit a provider (or the whole block) to run without it. See the settings reference for the full shape.

Install the runtimes

runtime_install builds/pulls the provider images for the enabled providers (it installs provider dependencies by default):

runtime_install()

This builds the images but does not start indexing on its own.

Start indexing and verify

Start the watchers so the providers index the configured repos and memory, then confirm they are healthy:

provider_watchers(action="start")   # start watchers
provider_watchers(action="refresh")  # re-seed after repo/memory changes
provider_status()                                   # compact readiness summary
provider_diagnostics()                              # raw provider-native detail

provider_watchers actions: status, start, stop, restart, refresh, shutdown-all. context_packet(repo_id=..., include_providers=true) also reports per-provider readiness, watcher state, and the target repo.

The c-13-install-and-onboard skill performs this start/verify step (stage 4) as part of first-run setup.

Use them

Once indexing, the search tools are available:

See the MCP tool reference for arguments.

Troubleshooting

Providers are optional: Agents Remember’s memory, onboarding, drift, and task workflows all work without them. They add faster code/memory search on top.