stash

<p align="center"> <a href="https://joinstash.ai"><img src="docs/assets/logo.svg" alt="Stash" width="320" /></a> </p> <h3 align="center">Help your agents learn from experience.</h3> <p align="center"> Agents generate valuable experience every time they work: successful approaches, <br> failed attempts, and human corrections. Stash captures that history and makes <br> its lessons available to future runs. </p> <p align="center"> <a href="https://github.com/Fergana-Labs/stash/actions/workflows/test.yml"><img src="https://github.com/Fergana-Labs/stash/actions/workflows/test.yml/badge.svg?branch=main" alt="CI" /></a> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-yellow.svg" alt="License: MIT" /></a> <a href="https://joinstash.ai"><img src="https://img.shields.io/badge/Website-joinstash.ai-F97316" alt="Website" /></a> <a href="https://discord.gg/PVFdcQx2u3"><img src="https://img.shields.io/badge/Discord-Join%20us-5865F2?logo=discord&logoColor=white" alt="Discord" /></a> </p> This repository provides the open-source foundation: trace collection, persistent knowledge, and reusable skills, accessible through a [Python SDK](sdk/README.md), REST API, MCP, and CLI. It works alongside your existing agents and models. ```mermaid flowchart LR A[Agent runs and human corrections] --> B[Capture sessions via hooks or API] B --> C[Curate durable knowledge] C --> D[Retrieve knowledge or package skills] D --> E[Use in the next agent run] E --> A ``` Stash's broader work focuses on extracting reliable feedback signals from messy production traces and using them to improve prompts, skills, and ultimately model weights. The [reward-model product docs](https://www.joinstash.ai/docs) cover trace review, reward-model training, and skill optimization with GEPA. The implementation is included in `backend/services/rm/` and `rm_worker/`. Our research focuses on improving the reliability of feedback extracted from production traces. The reward-model experiment is enabled for accounts created after the rollout. Existing accounts retain their current experience. Access is enforced in both the UI and API; see the [rollout guide](docs/reward-models/ROLLOUT.md). ## Reward models in action **Continuously improve prompts.** The Optimization tab connects a personal reward model to future agent runs through MCP. It proposes instructions, compares current and candidate prompts on real tasks, and tracks reward alongside a chosen business metric. See [continuous prompt optimization](docs/reward-models/CONTINUOUS-OPTIMIZATION.md) for setup, promotion criteria, pause/rollback and prototype limits. **Train from feedback in your traces.** Stash derives preference pairs from reviewer comments and user reactions, including approval, disappointment, and corrections. A classifier attributes each judgment to a response and keeps the source quote; unclear reactions are excluded from training. You can add comments by highlighting a response. There is no separate Auto mode or rating control in the UI; existing explicit ratings can be supplied through the API. Training needs at least two grounded comparisons. User reactions are used when available; an AI assessor also evaluates response quality without manual annotations or reactions. The [product demo](https://www.joinstash.ai/docs) below follows a refund-support agent from reviewer feedback to a reusable skill. These screenshots show an earlier interface with demo data; the current controls are described below. **1. Review a trace.** Highlight a response and explain what should change. Feedback already present in the conversation is extracted automatically during training. <!-- Frame from www/public/docs/demo/annotate.mp4 at 10 seconds. --> <p align="center"> <img src="docs/assets/reward-trace-feedback.png" alt="Refund-support trace with a highlighted refusal and reviewer feedback explaining how to handle a damaged order" width="900" /> </p> **2. Train a reward model from traces.** Select traces and choose **Create new reward model**. The worker extracts supported preferences, records their source, and trains the model. AI-generated comparisons receive a separate review for grounding and preference quality. **View learning** distinguishes user feedback from AI judgments, shows the supporting response or quote, and records whether each finding was included in training. AI judgments are model preferences, not measured customer satisfaction. <!-- Frame from www/public/docs/demo/train.mp4 at 3 seconds. --> <p align="center"> <img src="docs/assets/reward-model-training.png" alt="Six annotated traces selected for reward-model training, with three positive and three negative labels" width="900" /> </p> **3. Turn the reward signal into a skill.** Choose **View skill** on a trained model. It opens an existing run or starts one using GEPA, the selected traces, and their feedback. Download the resulting `SKILL.md` for your agent. <!-- Frame from www/public/docs/demo/skill.mp4 at 20 seconds. --> <p align="center"> <img src="docs/assets/reward-generated-skill.png" alt="Generated refund-requests skill showing candidate scores, reusable instructions, and a Download SKILL.md button" width="900" /> </p> [**Try Stash**](https://app.joinstash.ai) · [**Read the docs**](https://www.joinstash.ai/docs) For CLI setup, local development, and self-hosting, see [Running this repository](docs/running-stash.md). ## How it works 1. **Capture experience.** Hooks for coding agents record prompts, tool calls, and responses when session recording is enabled. Use the SDK or API to send events from your own agents. 2. **Extract durable lessons.** A scheduled curator reads new sessions and source material, then updates linked pages in your Memory wiki. The knowledge stays available after the original session ends. 3. **Make lessons reusable.** Agents search and read that knowledge through the CLI, MCP, API, or virtual filesystem. You and your agents can package related instructions and files into a Skill: a folder containing a `SKILL.md`. 4. **Carry them into future runs.** Install skills into your agent with `stash skills install`. Installed skills auto-update at session start, so changes to shared instructions can reach the next run without changing the underlying model's weights. ### Example: a correction becomes a reusable instruction Illustrative workflow: | Stage | What happens | |---|---| | **Trace** | An agent proposes a database migration. The reviewer points out that it would discard existing customer data. | | **Durable lesson** | Record the project rule: schema changes must migrate existing data forward. | | **Reusable skill** | Package a migration checklist that requires a data migration and verification that existing records survive. | | **Next run** | The agent loads the checklist while planning another schema change. Reviewers check whether it applied the lesson. | The output is inspectable knowledge and instructions that another agent can read, use, and revise. Whether they improve results should be checked on subsequent tasks. In an [internal experiment](https://henrydowling.com/agent-velocity.html), we measured a **49% speedup** for long-running Claude Code instances using Stash. See the experiment for its setup and results.