Skip to content
AI NEWS JOURNAL
Hacker News Show HNAI NEWS / JOURNAL

Show HN: Comemo – composite memory provider for Hermes Agent

There are many memory providers that can be used externally with Hermes Agent. However, like many other AI agents, Hermes only provides one slot for an external memory provider.Comemo takes that slot and acts as a composite provider. How? It fans out memory writes to multiple backends such as Mem0, Honcho, and Supermemory. On retrieval, it queries the enabled providers, merges their results, and removes duplicates before returning them to Hermes.I intend to add support for more memory providers and eventually extend Comemo to other open-source AI agents with similar limitations. Comments URL: https://news.ycombinator.com/item?id=50518538 Points: 1 # Comments: 0

Comemo is a composite memory provider for Hermes Agent. It runs multiple memory providers at the same time instead of one. Hermes lets you pick one memory provider. You can use Honcho, Mem0, Supermemory, or Hindsight, but only one at a time. Comemo sits in that single slot and writes to all of them at once. Reads come back merged and deduplicated. If one provider fails, the others keep running. The problem The community built several memory providers for Hermes: Honcho - cloud memory that tracks who you are across sessions Mem0 - reads every conversation and extracts facts automatically Supermemory - structured knowledge storage with semantic search Hermes only activates one at a time. This is a known constraint (Issue #24770, marked wontfix in core). The maintainers asked the community to build a composite provider as a catalog plugin instead. Comemo is that plugin. How it works Your conversation | v [ Hermes ] | v [ Comemo ] <-- implements MemoryProvider | +---> Honcho (optional, cloud) +---> Mem0 (optional, cloud) +---> Supermemory (optional, cloud) +---> Hindsight (optional, cloud) On every write, Comemo fans out to all active providers in parallel using a thread pool. If a provider fails, the write goes into a local SQLite retry queue and replays at the next session start. On every read, Comemo queries all active providers in parallel, merges the results, and deduplicates identical lines before returning to Hermes. There are two install paths. Use the one that matches how you got Hermes. From the Hermes marketplace hermes plugins install comemo Then run hermes memory setup — Hermes walks through enabling each provider and stores API keys securely in its secret scope. Restart Hermes after and Comemo is live. From source (GitHub) hermes plugins install andausman/comemo hermes memory setup After install: run hermes memory setup — Hermes presents each field from the config schema, writes non-secret settings to ~/.hermes/config.yaml, and stores API keys securely in its secret scope. hermes comemo setup Comemo — Composite Memory Meta-Provider ========================================== Comemo multiplexes multiple Hermes memory providers simultaneously. Each provider you enable runs in parallel — reads and writes fan out to all of them. Honcho (peer-based persistent user modeling) Tracks who you are across sessions. Get a key at honcho.dev Enable Honcho? [y/N]: y Workspace [hermes]: Mem0 (fact extraction via LLM + semantic retrieval) Automatically pulls facts from conversations. Get a key at mem0.ai Enable Mem0? [y/N]: y User ID [hermes-user]: Supermemory (3-layer retrieval: profile + entity graph + document RAG) Get a key at console.supermemory.ai Enable Supermemory? [y/N]: n Config written to ~/.hermes/config.yaml Active providers: honcho, mem0 Run 'hermes memory setup' to set your API keys, then start a new session. Commands Run these commands from your terminal once Comemo is installed and active. Runs the setup wizard. Prompts for which providers to enable and writes non-secret settings to ~/.hermes/config.yaml. Run hermes memory setup separately to set API keys. Shows what is configured without making any network calls. comemo (/home/user/.hermes/config.yaml) honcho enabled workspace=hermes mem0 enabled user_id=hermes-user supermemory disabled Connects to each enabled provider and does a real write and read. Confirms the providers are reachable and working, not just configured. testing Comemo providers (session: comemo-test-session-a1b2c3) honcho OK write + recall (found: TestUser, cross-session) mem0 OK write + recall (found: TestUser, comemo) supermemory SKIP not enabled or unavailable Queries all enabled providers and shows what is stored. Uses a broad recall by default so you get recent memories without needing to know what to search for. hermes comemo inspect inspecting Comemo providers (query: (all memories), limit: 20) honcho (3 entries) [1] User: my name is Anda, i'm building a Hermes plugin [2] Assistant: Got it, I'll remember that... [3] User: use Honcho for identity tracking mem0 (2 entries) [1] Anda is building a Hermes plugin for the Nous Research hackathon [2] Prefers cloud plus local hybrid memory setups supermemory SKIP not enabled or unavailable Use --query to filter by topic and --limit to control how many entries are shown per provider (default 20): hermes comemo inspect --query "python preferences" --limit 5 Verifying it works end to end Run hermes comemo test — all enabled providers should show OK. Start a Hermes session and have a short conversation. Mention something specific. Run hermes comemo inspect — you should see the conversation stored across your enabled providers. Start a new Hermes session and ask "what do you remember about me?" — Comemo injects the recalled memories into the system prompt, so Hermes will reference what was said before. Providers Cloud memory that understands who you are over time, not just what was said. Tracks identity and context across sessions using a workspace-scoped store. Requires an API key from honcho.dev. Set via HONCHO_API_KEY. Reads every conversation and automatically extracts facts like preferences, context, and recurring topics. Retrieves the most relevant ones at the start of each turn. Requires an API key from mem0.ai. Set via MEM0_API_KEY. Supermemory Structured knowledge storage with 3-layer retrieval: user profile, entity graph, and document RAG. Memories are scoped to a namespace. Requires an API key from supermemory.ai. Set via SUPERMEMORY_API_KEY. Hindsight Vectorized knowledge graph with entity resolution. Stores memories as interconnected entities and uses hybrid (vector + graph) retrieval for precise recall. Requires an API key from hindsight.vectorize.io. Set via HINDSIGHT_API_KEY. Optionally set a bank_id to scope memories per project or profile. Combinations You can enable any subset of providers: When Comemo is active it replaces the Hermes built-in memory provider. Existing built-in memories are not migrated automatically. Manual config If you prefer to configure without the wizard, add this to ~/.hermes/config.yaml: memory: provider: comemo plugins: comemo: honcho: enabled: true workspace: hermes mem0: enabled: true user_id: hermes-user supermemory: enabled: false namespace: hermes hindsight: enabled: false bank_id: "" api_url: https://api.hindsight.vectorize.io API keys are stored securely by Hermes via hermes memory setup — never written to config.yaml. Comemo reads them through the Hermes secret scope (HONCHO_API_KEY, MEM0_API_KEY, SUPERMEMORY_API_KEY, HINDSIGHT_API_KEY). Failure handling If a provider goes down, Comemo keeps running on the remaining ones. Failed writes go into a local SQLite retry queue. At the start of the next session, Comemo replays them automatically. Each item is retried up to 5 times.

Sources

Read original ↗

Related topics

← Continue to all updates