Integration

Long-term memory for Paperclip

Paperclip agents wake on heartbeats and lose accumulated context between runs. The plugin closes the gap in two ways: an event-driven worker auto-archives every issue, agent run, and approval into Engram (no agent code or LLM tool calls required), and four namespaced tools let agents read and write custom facts mid-run.

Install

Three steps: sign up for an Engram API key, paste a BYOK LLM-provider key on /models, then drop the snippet below into Paperclip.

Three steps to memory in your agent

  1. Sign up. Free, no card. You'll land on a Getting Started page that walks the next two steps.
  2. Add your LLM key. Engram is BYOK. Paste an OpenAI / Anthropic / Groq / Together / Fireworks key and we'll route every extraction and query call through your provider. You pay your provider directly. We never see your inference.
  3. Paste the snippet below into your agent and restart it. Use Authorization: Bearer <api-key>with the API key from your portal.

@lumetra/engram-paperclip-plugin: Paperclip plugin

Auto-archives every issue.created, agent.run.finished, and approval.decided event into Engram. No agent code, no LLM tool calls required. The Paperclip host emits domain events; the plugin's worker writes summaries to Engram automatically, giving you a permanent queryable record of the org's work from day one. Source: github.com/lumetra-io/engram-paperclip-plugin · npm: @lumetra/engram-paperclip-plugin.

  1. Install in a running Paperclip instance:
  2. Terminal
    paperclipai plugin install @lumetra/engram-paperclip-plugin
  3. Open the plugin settings at /instance/settings/plugins/io.lumetra.engram and paste your Engram API key, or set ENGRAM_API_KEY in your Paperclip server's environment and leave the form blank.
  4. Buckets are auto-created on first write, scoped per company by default (paperclip-company-<id>). Switch to per-project, per-agent, or global in the same settings form if you want a different shape.

What you can do once memory's wired in

  • Get a permanent, queryable record of every decision the agency made from day one
  • Auto-archive `issue.created` / `agent.run.finished` / `approval.decided` events with zero agent-side wiring
  • Scope memory per company, per project, per agent, or globally. The bucket strategy is a settings-form toggle.
  • Surface an `engram-stats` dashboard widget so operators see memory volume at a glance

FAQ

What's the difference between the event-archive path and the tool path?

Event-archive is passive: the Paperclip host emits domain events and the plugin worker writes summaries to Engram. No LLM in the loop, works on every adapter. The tool path lets agents call memory tools mid-run; it works via `tools/execute` REST today, and full MCP-tool surfacing depends on adapter support.

Does the `claude_local` adapter expose the memory tools to Claude?

Not yet. Paperclip surfaces only Claude CLI's own MCP servers there. The event-archive path is the working surface today; this one is tracked upstream.

Where does the plugin store its config?

In Paperclip's plugin settings, accessible at `/instance/settings/plugins/io.lumetra.engram`. The API key can also come from `ENGRAM_API_KEY` in the Paperclip server's env if you'd rather not persist it in plugin config.

Ship durable memory in Paperclip today

Free tier: 10K memories and 50K retrievals per month. No credit card. Same Engram backend powers all 41 integrations, so memories you write from one client are immediately queryable from the rest.