CognitiveXCOGNITIVEX · INTEGRATION
Persistent memory for Codex
Give the OpenAI Codex CLI long-term memory over MCP. Recall what you decided, store what you learned, across every session and repo. Hosted, nothing to run locally.
THE PROBLEM
Codex forgets the moment the session ends.
The OpenAI Codex CLI is a strong coding agent, but each run starts cold. Whatever you explained last week (why the auth layer is shaped the way it is, which migration broke staging, the convention you settled on for error handling) is gone when the conversation closes. You re-explain it. Then you re-explain it again.
CognitiveX gives Codex a memory that survives the session. It runs as an MCP server, so Codex can call remember to store a decision and recall to pull it back later, in a different session, a different terminal, or a different repo. The memory is hosted, it consolidates over time instead of growing without bound, and it follows your key everywhere. This is the same memory backbone behind the LCM, the Large Cognition Model. For Codex, it is the difference between an assistant that resets and one that builds on what it already knows about your codebase.
- Cross-session recall
- Works across repos
- Hosted over MCP
- Consolidates nightly
INSTALL IN ABOUT A MINUTE
Connect Codex to CognitiveX
CognitiveX is hosted at api.cognitivx.io. There is nothing to install locally. You point the Codex CLI at the endpoint and authenticate with a key.
1 · Create an API key. Sign in and generate a key from your CognitiveX developer dashboard.
2 · Add CognitiveX as an MCP server. The Codex CLI reads MCP servers from ~/.codex/config.toml. Add the CognitiveX endpoint and your key:
# ~/.codex/config.toml
[mcp_servers.cognitivx]
url = "https://api.cognitivx.io/mcp"
http_headers = { Authorization = "Bearer YOUR_API_KEY" }3 · Start Codex and let it remember. On your next run, Codex sees the CognitiveX tools. Ask it to recall context at the start of a task and to record decisions as it goes:
That is the whole setup. remember and recallare now part of Codex's tool set, and the memory consolidates server-side: pattern detection, salience scoring, and overnight dream consolidation run on the stored memories without any work on your side. Full reference is in the CognitiveX docs.
WHAT YOU GET
AGENTS.md vs. CognitiveX memory
Codex already supports a static context file. The two are complementary, but they solve different problems.
| Capability | AGENTS.md | Built-in session | CognitiveX |
|---|---|---|---|
| Survives across sessions | manual | ||
| Works across repos | |||
| Queried on demand (not pasted in full) | in-context | ||
| Learns & consolidates over time | |||
| Four memory tiers + salience + dream consolidation | |||
| Shared with your other agents & MCP clients |
AGENTS.md is the right home for fixed project rules. CognitiveX is the right home for everything you accumulate. See the full breakdown on the compare page.
NOT ANOTHER WRAPPER
The memory is the model.
An LLM does query → model → response → forget. CognitiveX closes the loop: query → living memory → reasoning → learning → evolution. What Codex stores does not sit in a flat log. It lands in one of four tiers: semantic (facts and architecture), episodic (what happened and when), procedural (how-tos and patterns), and foundational (the durable context that frames everything else). Salience scoring keeps the signal, pattern detection links related memories, and overnight consolidation compresses the noise into something worth recalling.
That structure is the point: CognitiveX is the cognitive infrastructure that makes every LLM smarter, not a thin cache bolted onto one. Read more about the Large Cognition Model, or try the consumer app, iCog, which is built on the same backbone.
QUESTIONS
Codex memory, answered.
Does the Codex CLI support MCP servers?
Yes. The Codex CLI reads MCP servers from its config (the [mcp_servers] table in ~/.codex/config.toml). Once CognitiveX is listed there, its memory tools (remember and recall) become callable by Codex during a session, the same way any other MCP tool is.
How is this different from AGENTS.md?
AGENTS.md is a static file you write and maintain by hand; it ships the same context into every run and never learns. CognitiveX is a living memory that persists across sessions and repos, consolidates what matters overnight, and is queried on demand instead of pasted in full. Use both: AGENTS.md for fixed project rules, CognitiveX for everything you accumulate over time.
Is there anything to run locally?
No. CognitiveX is hosted at api.cognitivx.io. You add one MCP entry pointing at the endpoint with a bearer token. There is no daemon, no local database, and no vector store to operate.
Does the memory work across projects?
Yes. The same memory backbone follows your key across every repo and every MCP client (Codex CLI, Claude Code, your own agents), so a decision recorded in one project can surface in another.
GET STARTED
Give Codex a memory that learns.
Hosted, consolidating, cross-session, installed over MCP in about a minute. Browse the other integrations to wire the same memory into the rest of your stack.