7f3dbaeb25
## Summary - enforce a 10 MiB cap per `thread_id` in state log storage - enforce a 10 MiB cap per `process_uuid` for threadless (`thread_id IS NULL`) logs - scope pruning to only keys affected by the current insert batch - add a cheap per-key `SUM(...)` precheck so windowed prune queries only run for keys that are currently over the cap - add SQLite indexes used by the pruning queries - add focused runtime tests covering both pruning behaviors ## Why This keeps log growth bounded by the intended partition semantics while preserving a small, readable implementation localized to the existing insert path. ## Local Latency Snapshot (No Truncation-Pressure Run) Collected from session `019c734f-1d16-7002-9e00-c966c9fbbcae` using local-only (uncommitted) instrumentation, while not specifically benchmarking the truncation-heavy regime. ### Percentiles By Query (ms) | query | count | p50 | p90 | p95 | p99 | max | |---|---:|---:|---:|---:|---:|---:| | `insert_logs.insert_batch` | 110 | 0.332 | 0.999 | 1.811 | 2.978 | 3.493 | | `insert_logs.precheck.process` | 106 | 0.074 | 0.152 | 0.206 | 0.258 | 0.426 | | `insert_logs.precheck.thread` | 73 | 0.118 | 0.206 | 0.253 | 1.025 | 1.025 | | `insert_logs.prune.process` | 58 | 0.291 | 0.576 | 0.607 | 1.088 | 1.088 | | `insert_logs.prune.thread` | 44 | 0.318 | 0.467 | 0.728 | 0.797 | 0.797 | | `insert_logs.prune_total` | 110 | 0.488 | 0.976 | 1.237 | 1.593 | 1.684 | | `insert_logs.total` | 110 | 1.315 | 2.889 | 3.623 | 5.739 | 5.961 | | `insert_logs.tx_begin` | 110 | 0.133 | 0.235 | 0.282 | 0.412 | 0.546 | | `insert_logs.tx_commit` | 110 | 0.259 | 0.689 | 0.772 | 1.065 | 1.080 | ### `insert_logs.total` Histogram (ms) | bucket | count | |---|---:| | `<= 0.100` | 0 | | `<= 0.250` | 0 | | `<= 0.500` | 7 | | `<= 1.000` | 33 | | `<= 2.000` | 40 | | `<= 5.000` | 28 | | `<= 10.000` | 2 | | `<= 20.000` | 0 | | `<= 50.000` | 0 | | `<= 100.000` | 0 | | `> 100.000` | 0 | ## Local Latency Snapshot (Truncation-Heavy / Cap-Hit Regime) Collected from a run where cap-hit behavior was frequent (`135/180` insert calls), using local-only (uncommitted) instrumentation and a temporary local cap of `10_000` bytes for stress testing (not the merged `10 MiB` cap). ### Percentiles By Query (ms) | query | count | p50 | p90 | p95 | p99 | max | |---|---:|---:|---:|---:|---:|---:| | `insert_logs.insert_batch` | 180 | 0.524 | 1.645 | 2.163 | 3.424 | 3.777 | | `insert_logs.precheck.process` | 171 | 0.086 | 0.235 | 0.373 | 0.758 | 1.147 | | `insert_logs.precheck.thread` | 100 | 0.105 | 0.251 | 0.291 | 1.176 | 1.622 | | `insert_logs.prune.process` | 109 | 0.386 | 0.839 | 1.146 | 1.548 | 2.588 | | `insert_logs.prune.thread` | 56 | 0.253 | 0.550 | 1.148 | 2.484 | 2.484 | | `insert_logs.prune_total` | 180 | 0.511 | 1.221 | 1.695 | 4.548 | 5.512 | | `insert_logs.total` | 180 | 1.631 | 3.902 | 5.103 | 8.901 | 9.095 | | `insert_logs.total_cap_hit` | 135 | 1.876 | 4.501 | 5.547 | 8.902 | 9.096 | | `insert_logs.total_no_cap_hit` | 45 | 0.520 | 1.700 | 2.079 | 3.294 | 3.294 | | `insert_logs.tx_begin` | 180 | 0.109 | 0.253 | 0.287 | 1.088 | 1.406 | | `insert_logs.tx_commit` | 180 | 0.267 | 0.813 | 1.170 | 2.497 | 2.574 | ### `insert_logs.total` Histogram (ms) | bucket | count | |---|---:| | `<= 0.100` | 0 | | `<= 0.250` | 0 | | `<= 0.500` | 16 | | `<= 1.000` | 39 | | `<= 2.000` | 60 | | `<= 5.000` | 54 | | `<= 10.000` | 11 | | `<= 20.000` | 0 | | `<= 50.000` | 0 | | `<= 100.000` | 0 | | `> 100.000` | 0 | ### `insert_logs.total` Histogram When Cap Was Hit (ms) | bucket | count | |---|---:| | `<= 0.100` | 0 | | `<= 0.250` | 0 | | `<= 0.500` | 0 | | `<= 1.000` | 22 | | `<= 2.000` | 51 | | `<= 5.000` | 51 | | `<= 10.000` | 11 | | `<= 20.000` | 0 | | `<= 50.000` | 0 | | `<= 100.000` | 0 | | `> 100.000` | 0 | ### Performance Takeaways - Even in a cap-hit-heavy run (`75%` cap-hit calls), `insert_logs.total` stays sub-10ms at p99 (`8.901ms`) and max (`9.095ms`). - Calls that did **not** hit the cap are materially cheaper (`insert_logs.total_no_cap_hit` p95 `2.079ms`) than cap-hit calls (`insert_logs.total_cap_hit` p95 `5.547ms`). - Compared to the earlier non-truncation-pressure run, overall `insert_logs.total` rose from p95 `3.623ms` to p95 `5.103ms` (+`1.48ms`), indicating bounded overhead when pruning is active. - This truncation-heavy run used an intentionally low local cap for stress testing; with the real 10 MiB cap, cap-hit frequency should be much lower in normal sessions. ## Testing - `just fmt` (in `codex-rs`) - `cargo test -p codex-state` (in `codex-rs`) |
||
|---|---|---|
| .. | ||
| .cargo | ||
| .config | ||
| .github/workflows | ||
| ansi-escape | ||
| app-server | ||
| app-server-protocol | ||
| app-server-test-client | ||
| apply-patch | ||
| arg0 | ||
| async-utils | ||
| backend-client | ||
| chatgpt | ||
| cli | ||
| cloud-requirements | ||
| cloud-tasks | ||
| cloud-tasks-client | ||
| codex-api | ||
| codex-backend-openapi-models | ||
| codex-client | ||
| codex-experimental-api-macros | ||
| config | ||
| core | ||
| debug-client | ||
| docs | ||
| exec | ||
| exec-server | ||
| execpolicy | ||
| execpolicy-legacy | ||
| feedback | ||
| file-search | ||
| hooks | ||
| keyring-store | ||
| linux-sandbox | ||
| lmstudio | ||
| login | ||
| mcp-server | ||
| network-proxy | ||
| ollama | ||
| otel | ||
| process-hardening | ||
| protocol | ||
| responses-api-proxy | ||
| rmcp-client | ||
| scripts | ||
| secrets | ||
| shell-command | ||
| state | ||
| stdio-to-uds | ||
| tui | ||
| utils | ||
| vendor | ||
| windows-sandbox-rs | ||
| .gitignore | ||
| BUILD.bazel | ||
| Cargo.lock | ||
| Cargo.toml | ||
| clippy.toml | ||
| code | ||
| config.md | ||
| default.nix | ||
| deny.toml | ||
| node-version.txt | ||
| README.md | ||
| rust-toolchain.toml | ||
| rustfmt.toml | ||
Codex CLI (Rust Implementation)
We provide Codex CLI as a standalone, native executable to ensure a zero-dependency install.
Installing Codex
Today, the easiest way to install Codex is via npm:
npm i -g @openai/codex
codex
You can also install via Homebrew (brew install --cask codex) or download a platform-specific release directly from our GitHub Releases.
Documentation quickstart
- First run with Codex? Start with
docs/getting-started.md(links to the walkthrough for prompts, keyboard shortcuts, and session management). - Want deeper control? See
docs/config.mdanddocs/install.md.
What's new in the Rust CLI
The Rust implementation is now the maintained Codex CLI and serves as the default experience. It includes a number of features that the legacy TypeScript CLI never supported.
Config
Codex supports a rich set of configuration options. Note that the Rust CLI uses config.toml instead of config.json. See docs/config.md for details.
Model Context Protocol Support
MCP client
Codex CLI functions as an MCP client that allows the Codex CLI and IDE extension to connect to MCP servers on startup. See the configuration documentation for details.
MCP server (experimental)
Codex can be launched as an MCP server by running codex mcp-server. This allows other MCP clients to use Codex as a tool for another agent.
Use the @modelcontextprotocol/inspector to try it out:
npx @modelcontextprotocol/inspector codex mcp-server
Use codex mcp to add/list/get/remove MCP server launchers defined in config.toml, and codex mcp-server to run the MCP server directly.
Notifications
You can enable notifications by configuring a script that is run whenever the agent finishes a turn. The notify documentation includes a detailed example that explains how to get desktop notifications via terminal-notifier on macOS. When Codex detects that it is running under WSL 2 inside Windows Terminal (WT_SESSION is set), the TUI automatically falls back to native Windows toast notifications so approval prompts and completed turns surface even though Windows Terminal does not implement OSC 9.
codex exec to run Codex programmatically/non-interactively
To run Codex non-interactively, run codex exec PROMPT (you can also pass the prompt via stdin) and Codex will work on your task until it decides that it is done and exits. Output is printed to the terminal directly. You can set the RUST_LOG environment variable to see more about what's going on.
Use codex exec --ephemeral ... to run without persisting session rollout files to disk.
Experimenting with the Codex Sandbox
To test to see what happens when a command is run under the sandbox provided by Codex, we provide the following subcommands in Codex CLI:
# macOS
codex sandbox macos [--full-auto] [--log-denials] [COMMAND]...
# Linux
codex sandbox linux [--full-auto] [COMMAND]...
# Windows
codex sandbox windows [--full-auto] [COMMAND]...
# Legacy aliases
codex debug seatbelt [--full-auto] [--log-denials] [COMMAND]...
codex debug landlock [--full-auto] [COMMAND]...
Selecting a sandbox policy via --sandbox
The Rust CLI exposes a dedicated --sandbox (-s) flag that lets you pick the sandbox policy without having to reach for the generic -c/--config option:
# Run Codex with the default, read-only sandbox
codex --sandbox read-only
# Allow the agent to write within the current workspace while still blocking network access
codex --sandbox workspace-write
# Danger! Disable sandboxing entirely (only do this if you are already running in a container or other isolated env)
codex --sandbox danger-full-access
The same setting can be persisted in ~/.codex/config.toml via the top-level sandbox_mode = "MODE" key, e.g. sandbox_mode = "workspace-write".
Code Organization
This folder is the root of a Cargo workspace. It contains quite a bit of experimental code, but here are the key crates:
core/contains the business logic for Codex. Ultimately, we hope this to be a library crate that is generally useful for building other Rust/native applications that use Codex.exec/"headless" CLI for use in automation.tui/CLI that launches a fullscreen TUI built with Ratatui.cli/CLI multitool that provides the aforementioned CLIs via subcommands.
If you want to contribute or inspect behavior in detail, start by reading the module-level README.md files under each crate and run the project workspace from the top-level codex-rs directory so shared config, features, and build scripts stay aligned.