fleet-memory/hindsight-integrations/codex/README.md
Ben 0b17a67c70
feat: add Hindsight memory integration for OpenAI Codex CLI (#730)
* feat(codex): add Hindsight memory integration for OpenAI Codex CLI

Hooks-based integration that gives Codex CLI long-term memory via Hindsight.
Three hooks keep memory in sync: SessionStart (daemon pre-warm), UserPromptSubmit
(recall + context injection), Stop (retain conversation to memory).

Key differences from the Claude Code integration:
- Codex transcript format: JSONL with {msg: {type, message}} (user_message/agent_message)
- No CODEX_PLUGIN_ROOT env var — install.sh writes hooks.json with absolute paths
- State stored in ~/.hindsight/codex/state/ (not CLAUDE_PLUGIN_DATA)
- No async: true in hooks (not supported by Codex)
- No SessionEnd event
- hooks.json written to ~/.codex/hooks.json with codex_hooks = true in config.toml

* fix(codex): fix transcript parser for actual Codex disk format

Codex stores sessions as rollout-*.jsonl with response_item entries:
  User:      {type:response_item, payload:{type:message, role:user, content:[{type:input_text, text:...}]}}
  Assistant: {type:response_item, payload:{type:message, role:assistant, phase:final_answer, content:[{type:output_text, text:...}]}}

Previous parser expected an undocumented {msg:{type:user_message}} format from the Rust protocol spec
that does not match the actual on-disk storage format.

* feat(codex): add reflect mode to UserPromptSubmit hook

Add recallMode config option (default: 'recall') that switches the
UserPromptSubmit hook between:
- 'recall': existing behavior, fast raw facts list
- 'reflect': agentic synthesis loop, returns coherent prose answer

Also adds reflect() method to HindsightClient and HINDSIGHT_RECALL_MODE
env var override. Reflect uses a 25s timeout (vs 10s for recall).

* feat(codex): auto mode for recall/reflect selection

Add recallMode: 'auto' (new default) that picks the operation per-query:
- Synthesis patterns (what do you know, what's my, summarize, etc.) → reflect
- All other prompts → recall (fast, raw facts, better for code tasks)

* feat(codex): add automated test suite and finalize recall-only mode

* docs(codex): add docs page and sidebar entry for Codex CLI integration
2026-03-30 10:51:53 +02:00

127 lines
4 KiB
Markdown

# Hindsight for OpenAI Codex CLI
Long-term memory for [OpenAI Codex CLI](https://github.com/openai/codex) — remembers your projects, preferences, and past sessions across every conversation.
## How it works
Three Codex hooks keep memory in sync automatically:
| Hook | Action |
|------|--------|
| `SessionStart` | Warms up the Hindsight server in the background |
| `UserPromptSubmit` | Recalls relevant memories and injects them into context |
| `Stop` | Retains the conversation to long-term memory |
## Requirements
- **OpenAI Codex CLI** v0.116.0 or later (hooks support)
- **Python 3.9+** (for hook scripts)
- **Hindsight**: [Hindsight Cloud](https://hindsight.vectorize.io) or local `hindsight-embed`
## Installation
```bash
git clone https://github.com/vectorize-io/hindsight
cd hindsight/hindsight-integrations/codex
./install.sh
```
The installer:
1. Copies scripts to `~/.hindsight/codex/scripts/`
2. Writes `~/.codex/hooks.json` with absolute paths to the scripts
3. Adds `codex_hooks = true` to `~/.codex/config.toml`
### Uninstall
```bash
./install.sh --uninstall
```
## Configuration
The default config is written to `~/.hindsight/codex/settings.json` on first install.
For personal overrides (stable across updates), create `~/.hindsight/codex.json`:
```json
{
"hindsightApiUrl": "https://api.hindsight.vectorize.io",
"hindsightApiToken": "your-api-key",
"bankId": "my-codex-memory"
}
```
### Hindsight Cloud
```json
{
"hindsightApiUrl": "https://api.hindsight.vectorize.io",
"hindsightApiToken": "your-api-key"
}
```
### Local daemon (hindsight-embed)
Set an LLM API key and Hindsight will start the local server automatically:
```bash
export OPENAI_API_KEY=sk-your-key
# or
export ANTHROPIC_API_KEY=your-key
```
### Configuration options
| Key | Default | Description |
|-----|---------|-------------|
| `hindsightApiUrl` | `""` | External API URL (empty = local daemon) |
| `hindsightApiToken` | `null` | API token for Hindsight Cloud |
| `bankId` | `"codex"` | Memory bank identifier |
| `bankMission` | (set) | Guides what facts Hindsight retains |
| `autoRecall` | `true` | Inject memories before each prompt |
| `autoRetain` | `true` | Store conversations after each turn |
| `retainMode` | `"full-session"` | `"full-session"` or `"chunked"` |
| `retainEveryNTurns` | `10` | Retain every N turns (1 = every turn) |
| `recallBudget` | `"mid"` | Recall depth: `"low"`, `"mid"`, `"high"` |
| `recallMaxTokens` | `1024` | Max tokens for injected memories |
| `dynamicBankId` | `false` | Separate bank per project/session |
| `dynamicBankGranularity` | `["agent", "project"]` | Fields for dynamic bank ID |
| `debug` | `false` | Log debug info to stderr |
### Environment variable overrides
All settings can also be set via environment variables:
```bash
export HINDSIGHT_API_URL=https://api.hindsight.vectorize.io
export HINDSIGHT_API_TOKEN=your-api-key
export HINDSIGHT_BANK_ID=my-project
export HINDSIGHT_DEBUG=true
```
## How memory works
**Recall** — before each prompt, Hindsight searches your memory bank for facts relevant to what you're about to ask. Found memories are injected as context so Codex has continuity across sessions.
**Retain** — after each turn, Codex's conversation is stored to Hindsight. The memory engine extracts facts, relationships, and experiences — so you don't need to re-explain your stack, preferences, or past decisions.
## Dynamic bank IDs
To keep separate memory per project:
```json
{
"dynamicBankId": true,
"dynamicBankGranularity": ["agent", "project"]
}
```
This creates banks like `codex::my-project` automatically, using the working directory name.
## Troubleshooting
**Memory not appearing**: Enable debug mode (`"debug": true`) and check stderr output.
**Server not starting**: Set `hindsightApiUrl` to use an external server, or ensure `uvx` is on PATH for local daemon mode.
**Hooks not firing**: Check that `~/.codex/config.toml` contains `codex_hooks = true` under `[features]`, and that your Codex CLI version supports hooks (v0.116.0+).