Installation Overview
Hippocortex integrates into your AI stack through three primary paths. Each one gives your agent persistent memory that learns from experience. Pick the path that matches your setup.
Choose Your Path
Gateway (Recommended)
Change your base URL. That is the entire integration.
The Gateway is an OpenAI-compatible proxy. Any application that uses the OpenAI SDK (Python or TypeScript) can add memory by pointing at the Hippocortex endpoint instead of the provider's endpoint. No SDK to install, no code to change.
- Setup time: 2 minutes
- Reliability: ~99% (fully automatic with graceful fallback)
- Streaming: Yes
- Auto-capture: Yes
- Auto-compile: Yes
- Full pipeline: Semantic search, graph retrieval, collective brain, behavioral context, vault
- Works with: OpenAI, Anthropic, Google Gemini, Groq, Together, Mistral, Fireworks, Ollama, and any OpenAI-compatible provider
SDK
Install the npm or PyPI package. One import or one wrap() call adds memory to every LLM call.
The SDK gives you more control than the Gateway. You can choose which clients get memory, capture custom events (file edits, tool calls, commands), and call synthesize() independently of LLM calls.
- Setup time: 5 minutes
- Reliability: ~95% (auto-instrumentation)
- Streaming: Yes
- Auto-capture: With auto-instrumentation or wrap()
- Auto-compile: Yes
- Full pipeline: Same as Gateway
- Works with: OpenAI, Anthropic (native SDK support)
OpenClaw Plugin (Recommended for OpenClaw)
Add three lines to your OpenClaw config. Your agent remembers everything across sessions.
The plugin hooks into OpenClaw's lifecycle to capture every conversation and inject relevant context automatically. Includes the Context Engine for infinite session management. No code changes, no manual memory management.
- Setup time: 3 minutes
- Reliability: ~95% (fully automatic)
- Auto-capture: Yes
- Auto-compile: Yes
- Full pipeline: Same as Gateway + Context Engine for infinite sessions
- Works with: Any OpenClaw agent
Comparison Matrix
| Gateway | SDK | OpenClaw Plugin | |
|---|---|---|---|
| Setup time | 2 min | 5 min | 3 min |
| Code changes | None | 1 line | None |
| Reliability | ~99% | ~95% | ~95% |
| Auto-capture | Yes | Yes | Yes |
| Auto-compile | Yes | Yes | Yes |
| Streaming | Yes | Yes | N/A |
| Custom events | No | Yes | No |
| Vault | Yes | Yes | Yes |
| Collective Brain | Yes | Yes | Yes |
| Context Engine | No | No | Yes |
| Offline/self-hosted | Yes | Yes | Yes |
Framework Adapters (Python)
For teams using specific AI frameworks, Hippocortex provides Python adapters:
| Framework | Package | Status |
|---|---|---|
| OpenAI Agents | pip install hippocortex[openai-agents] | Stable |
| LangGraph | pip install hippocortex[langgraph] | Stable |
| CrewAI | pip install hippocortex[crewai] | Beta |
| AutoGen | pip install hippocortex[autogen] | Beta |
These adapters wrap your framework's execution to capture events and inject context automatically. See the Adapters documentation for details.
Which Path Should I Pick?
"I just want memory on my existing app." Use the Gateway. Change one URL, done.
"I want full control over what gets captured." Use the SDK. Wrap individual clients, capture custom events.
"I use OpenClaw." Use the OpenClaw Plugin. Three lines of config, fully automatic.
"I use Claude Code, Cursor, or Windsurf." Use the Gateway. Set your base URL to Hippocortex.
"I use multiple paths." They all work together. Gateway for your main app, SDK for custom scripts, Plugin for OpenClaw. All memory ends up in the same store.
Next Steps
Pick a path and follow the guide. Each one gets you from zero to working memory in under 5 minutes.
For step-by-step onboarding guides tailored to specific agents and frameworks, see the Onboarding section.