openreflex 0.3.2

Website · Docs · Walkthrough · PyPI · Issues · Buy me a coffee

What is OpenReflex

OpenReflex gives AI coding agents muscle memory. It plugs into Claude Code, Codex, Cursor, and OpenCode through lifecycle hooks and MCP, quietly records how each task actually went, and hands the next similar task what worked before: the approach, the files that mattered, and the fix for the error you hit last time.

You install it once and keep working normally. Everything stays on your machine in a local SQLite database. There is no account, no service, and no telemetry.

_{A real Claude Code session, recorded from the terminal and sped up. OpenReflex had seen one similar task in this repository, so it adds 162 tokens of context before the first tool call; after the fix, Claude records the outcome through OpenReflex's MCP tool. The step-by-step walkthrough is in docs/walkthrough.md.}

Why OpenReflex

• It remembers what worked. Before a substantial task, OpenReflex retrieves similar past tasks and injects a compact Execution Context: a suggested approach with alternatives, the files that were changed, and lessons such as which edit resolved a recurring error.
• It catches loops while they happen. Repeated failing commands, identical retries, stalled progress, and runaway context growth raise one alert that says whether to continue, pivot to another approach, or stop and check in with you, never a stream of nags.
• It learns after every task. OpenReflex infers the outcome from real verification (a test or build that passed or failed after the last edit), estimates Execution Regret against the alternatives, and extracts lessons.
• It is private by design. Only coarse, project-relative metadata is stored. File contents, commands, tool output, and transcripts never are, and capture is off until you approve a project.

Quick start

Requires Python 3.11 or newer.

pipx install openreflex            # or: uv tool install openreflex
cd your-project
openreflex install claude-code     # writes hooks + MCP config and enables this project

OPENREFLEX / CONNECT
  agent       claude-code
  project     D:\demo\shop
  config      updated 2 files
              D:\demo\shop\.claude\settings.json
              D:\demo\shop\.mcp.json
  memory      enabled
  storage     local

  [ok] reflex active

Then work as usual. After a few tasks:

openreflex context "fix the login redirect bug"   # preview the context a task would receive
openreflex status                                 # what has been captured and learned
openreflex doctor                                 # installation checks and recent hook errors

Use it with your agent

OpenReflex installs per project with openreflex install <agent>, or as a plugin.

With a plugin install, enable each project with openreflex approve. Per-agent guides: Claude Code, Codex, Cursor, OpenCode.

How it works

Hooks send lifecycle events (prompt, tool start, tool end, compaction, stop) to the OpenReflex engine, which stores them in an Experience Graph:

Task -caused-> Execution -used-> Context -used-> Experience
CandidatePath -recommended_for-> Task            Lesson -recommended_for-> Task
Execution -failed_with-> ToolCall -resolved_by-> ToolCall
Execution -caused-> Outcome -caused-> Experience -caused-> Lesson

• Before a task: similar experiences are retrieved and three strategies (inspect-first, test-first, incremental) are estimated on success probability, time, tool calls, context cost, risk, uncertainty, reversibility, and expected regret. Strategies that another one beats on every measure are dropped as dominated (Pareto efficiency), the rest are ranked by utility within any limits you set, and the chosen path gets an execution budget for time, tool calls, and context. A context of at most 1,400 characters is injected; nothing is injected without relevant experience.
• During a task: when a detector finds a failure loop, repeated calls, stalled progress, context growth, or work past the budget, OpenReflex estimates the marginal value of more work. The current path's success estimate is updated with each call that makes no progress or fails, and compared with the cost of the work left and with the best untried alternative. The alert ends with a recommendation to continue, pivot to another strategy, or stop and ask the user. Each problem, pivot, or stop is raised once, with a cooldown between messages.
• After a task: outcome and chosen path come from the agent's record_outcome / choose_path MCP calls when available, and are otherwise inferred from tool activity. Execution Regret compares the path taken with the best plausible alternative; it is withheld when the outcome is unknown, and feeds back into how strategies are ranked next time, along with success, cost, and how often a strategy ran into trouble.

Every recommendation is stored as a decision snapshot with a Reflex Score (0-100): how strong the recommendation is, which is separate from the estimated chance that the task succeeds. openreflex why explains the latest decision and openreflex trace shows the timeline. In Claude Code, a short recap appears when a task starts, when OpenReflex recommends a pivot or stop or detects trouble, and when the task completes.

Set optional limits for every task with OPENREFLEX_BUDGET, for example calls=40,minutes=20,tokens=60000. Routing, scoring, budget and recap settings come from a versioned policy: the packaged defaults, overridden by ~/.openreflex/config.toml and then by .openreflex.toml in the project.

Agents can also query OpenReflex directly through its MCP server.

CLI

Privacy

Data lives in ~/.openreflex/projects/<hash>/experience.sqlite3. Set OPENREFLEX_HOME to move it, OPENREFLEX_DISABLE=1 to turn capture off everywhere, openreflex forget --yes to delete a project's data, or ask your agent to forget_experience a single task.

Research

OpenReflex is also a research project in budget-aware execution: instead of treating success as a yes or no, it studies how agents choose execution paths, spend tool calls and context, respond to uncertainty, and whether more computation still adds value. The loop is experience retrieval, Pareto path selection, budget-aware execution and counterfactual regret. The Researcher view on the site explains the idea, lets you step through one reflex forming in the graph, and places it next to related work.

Community & Contributing

• Issues and ideas: GitHub Issues

• Support the project: Buy me a coffee

• Develop locally:

  git clone https://github.com/vishnu-77/openreflex && cd openreflex
  pip install -e ".[dev]"
  pytest && ruff check src tests scripts
  

OpenReflex is listed in the MCP Registry as io.github.vishnu-77/openreflex and on Glama:

License

OpenReflex is released under the MIT License.