# SpecStory: Local-First Session History and Knowledge Extraction for AI Coding Tools > SpecStory captures every conversation from AI coding assistants such as Cursor, Claude Code, and Copilot, saves them to .specstory/history/ in each project, and makes them searchable, resumable, and shareable. The /lore command converts accumulated history into evidence-backed agent skills. **specstoryai/getspecstory** — Install our local first extensions for your favorite AI IDE or Terminal Agent. Process your histories into reusable skills with Lore. Sync your conversations to the cloud. File issues and requests. - Repository: https://github.com/specstoryai/getspecstory - Website: https://specstory.com/ - Stars: 1,343 · Forks: 89 - Language: Go - License: Apache-2.0 - Published: 2026-09-15 · Updated: 2026-09-15 · Language: en - Canonical page: https://hysenlabs.com/projects/specstoryai-getspecstory ## The Problem SpecStory Addresses: Lost AI Development Context AI coding tools produce useful outputs: architectural decisions, working code snippets, debugging sessions that resolved a subtle bug, and step-by-step explanations of unfamiliar codebases. The standard experience is that these conversations live only in the tool's session window. When the session ends, the context is gone. A developer who wants to revisit a solution from last week has to either find it in the tool's own history (if the tool keeps one) or reconstruct the reasoning from scratch. SpecStory solves this by capturing every AI interaction as it happens and writing it to a local directory within the project. The README describes the value as ensuring that "no brilliant solution, code snippet, or architectural decision" is lost. The .specstory/history/ path is the central artifact: a structured record of every session that can be searched, browsed, and eventually processed into reusable form. The project is built in Go, licensed under Apache-2.0, and releases are frequent: v2.14.1, v2.15.0, and v2.15.1 all shipped within a three-day window in September 2026. ## How Local-First Capture Works Across Thirteen Tools SpecStory supports two categories of tool: IDE extensions and CLI agents. For IDE tools such as Cursor and GitHub Copilot, SpecStory ships as a VS Code-compatible extension that intercepts sessions from within the editor. For CLI-based agents including Claude Code, Codex CLI, Cursor CLI, Droid CLI, Gemini CLI, DeepSeek TUI, Antigravity CLI, Muse Code, and several others, SpecStory operates as a CLI tool that wraps the agent session. All captured sessions write to .specstory/history/ in the project root. The README describes this as the central local store. The format is local-first: data stays on the developer's machine unless the developer chooses to sync to SpecStory Cloud. The README's workflow diagram shows the path: AI tools on the left, the local history directory in the middle, and SpecStory Cloud as an optional right endpoint. The minimum version requirements per tool vary. Claude Code requires v1.0.27 or later. GitHub Copilot requires v1.300.0 or later. Cursor CLI requires v2025.09.18 or later. Each integration has its own minimum, and the README's full table lists the requirement for each supported agent. ## Installing SpecStory for CLI Agents IDE extensions are installed through the VS Code extension marketplace by searching for SpecStory in the Extensions panel. CLI agent coverage installs through Homebrew: ```bash brew tap specstoryai/tap brew install specstory ``` Once installed, the CLI wraps the agent session automatically. Sessions are saved to .specstory/history/ in the directory where the agent was started, which is typically the project root. The repository also includes an install.sh script at the root for non-Homebrew environments. The .claude-plugin/ directory in the repository provides a plugin integration for Claude Code specifically. The .specstory/ directory present in the repository itself is an example of the history directory structure: the project uses SpecStory for its own development. For Docker-based deployments, the repository includes a Dockerfile. The .goreleaser.yml file handles automated release builds across platforms, which explains the frequent release cadence. ## Lore: Turning Session History into Agent Skills Once a developer has accumulated sessions in .specstory/history/, the /lore command mines that history into reusable, evidence-backed skills for AI agents. The README describes Lore as a separate component under the lore/ directory in the repository. The mechanism is that Lore processes the captured conversation history, identifies recurring patterns, solutions, and decisions, and packages them as structured skills that can be injected into future agent prompts. The README describes the result as skills with evidence, meaning the skill includes references back to the sessions that produced it rather than presenting conclusions without supporting history. Lore is positioned as the long-term payoff of the history capture workflow. A developer who has been capturing sessions for weeks or months has a corpus of real solutions from their specific codebase. Lore converts that corpus from a searchable archive into active context that agents can use in new sessions without requiring the developer to manually locate and paste the relevant history. ## SpecStory Cloud: Search, Chat, and Team Sharing SpecStory Cloud at cloud.specstory.com is the optional hosted layer. The README lists four cloud-specific capabilities: searching across sessions from multiple machines, asking questions across all sessions to revisit past decisions, resuming sessions from a different machine or by a different team member, and sharing specific solutions with colleagues. The local capture workflow works without a SpecStory Cloud account. Cloud features require login. The README describes the sync as optional: developers who want to keep all history local can use SpecStory exclusively as a local tool and never sync. For teams, the shared session model means that when one engineer solves a problem in an AI session, the solution is available to teammates in the cloud search index rather than locked to one machine's local history directory. The README frames this as making AI-derived knowledge a team asset rather than an individual one. ## Limitations and When SpecStory Adds No Value SpecStory captures conversations but does not evaluate them. Every session goes into history regardless of whether the outcome was useful. A developer who uses AI tools heavily will accumulate a large history directory that includes both valuable sessions and sessions that led nowhere. Searching that history requires knowing what to look for, and Lore's skill extraction depends on there being enough signal in the history to identify patterns. The extension and CLI coverage is wide but not universal. The README documents thirteen tools by name, with the note that newer tools may not yet have integrations. Engineers using a tool not on the list will get no capture without writing a custom integration against SpecStory's open-source CLI. An alternative for teams using Cursor specifically is Cursor's own history feature, which keeps session context within the IDE. The difference is that Cursor's history is local to the IDE and not portable across tools or team members, while SpecStory writes a tool-agnostic local format and optionally syncs to a shared cloud index. Teams that use only one AI tool and have no need for cross-tool or cross-team session sharing may find Cursor's built-in history sufficient. ## Conclusion SpecStory is useful for any engineer who works daily with AI coding tools and has lost time reconstructing a solution they remember arriving at in a previous session. The local-first design means no data leaves the machine without explicit sync consent, and the history directory is a plain folder that can be opened, diffed, and committed to git. Teams that have never experienced the problem of losing a productive AI session may find the overhead of maintaining history not worth the setup cost. The /lore feature is only valuable once there is a substantial history to mine. Before adopting it, check that your specific AI coding tool version meets the minimum version listed in the README's compatibility table, since each tool integration has its own minimum version requirement. ## FAQ ### Does SpecStory send my code or sessions to any server by default? No. SpecStory is local-first: all captured sessions write to .specstory/history/ on the local machine. Data is sent to SpecStory Cloud only when the developer explicitly chooses to sync. Local search and session resumption work without any network connection. ### Which AI coding tools does SpecStory support? The README lists Cursor IDE, GitHub Copilot, Claude Code, Codex CLI, Cursor CLI, Droid CLI, Gemini CLI, DeepSeek TUI, Antigravity CLI, Muse Code, OpenCode, Pi, Qwen, and Grok Build. IDE tools are supported via a VS Code extension; CLI tools are supported via the SpecStory CLI installed through Homebrew. ### What does the SpecStory /lore command produce? Running /lore processes the accumulated session history in .specstory/history/ and extracts reusable, evidence-backed skills for AI agents. The skills reference the original sessions they were derived from, giving agents access to solutions that were discovered in real work on the same codebase. ## Sources - [License: Apache-2.0](https://github.com/specstoryai/getspecstory/blob/dev/LICENSE) - [Project website](https://specstory.com/) - [README](https://github.com/specstoryai/getspecstory/blob/dev/README.md) - [Releases](https://github.com/specstoryai/getspecstory/releases) - [specstoryai/getspecstory on GitHub](https://github.com/specstoryai/getspecstory) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/specstoryai-getspecstory