Kurrent, the company behind the event‑native database KurrentDB, has unveiled Kurrent Capacitor, an AI‑native shared‑memory platform designed specifically for coding agents such as Claude Code, Codex and Cursor. In a private‑preview launch announced in May 2024, Kurrent positioned Capacitor as the first solution that records every agent turn, test run, hypothesis, and reasoning block, then makes that immutable record searchable, shareable, and actionable for both human developers and future agent sessions. By streaming each event to a real‑time dashboard, a command‑line interface (CLI), and built‑in MCP tools, Capacitor promises to turn what were previously siloed transcripts into a collaborative knowledge base that can be queried, replayed, and extended across the entire software development lifecycle.
Kurrent Launches Capacitor Private Preview
The launch announcement frames Capacitor as a foundational layer for “humans and coding agents building software together.” Access to the preview is open immediately, with teams able to request entry at https://capacitor.kurrent.io/. Once granted, users receive a real‑time dashboard that visualizes session flow, a CLI (open‑source at https://github.com/kurrent-io/kcap-cli) for scriptable interaction, and MCP tools that agents themselves can call to retrieve prior context. Kurrent emphasizes that the platform “captures every coding agent session … and makes that record searchable, shareable, and actionable across the entire team and across future agent runs,” eliminating the need for separate record‑keeping commands or per‑session configuration.
CEO Kirk Dunn is quoted saying the tool “provides interaction between teams of agents and humans in real time. Collaborative, persistent, searchable and secure session memory from the first prompt to production.” The company notes that the service is free throughout the preview phase and works out‑of‑the‑box with Claude Code, Codex and Cursor, requiring no additional setup per session. This low barrier to entry is intended to accelerate adoption among teams that are already experimenting with agentic coding but struggling with lost context and fragmented handoffs.
Technical Architecture and Core Capabilities
Capacitor rests on KurrentDB, Kurrent’s event‑native database that stores each agent turn, tool call, and lifecycle event as an immutable entry in an append‑only stream. This architecture mirrors audit‑friendly patterns used in regulated sectors, delivering deterministic records that can be replayed, queried, and projected into custom views. Because the event schema is vendor‑neutral, sessions from Claude Code, Codex, Cursor, or any future agent can be serialized to the same stream without breaking continuity.
The platform delivers six core capabilities, each directly tied to challenges outlined in Kurrent’s source material:
- Collaboration – Live session links can be shared via Slack, Teams, or other collaboration tools, letting teammates view the evolving reasoning, add comments, or contribute code without having to reconstruct context from fragmented messages.
- Multi‑agent handoffs – Agents can pick up exactly where another left off, preserving full context and allowing model switches or parallel workstreams. This enables fluid handoffs, “handing back” and “continuing work” without losing any decision history.
- PR review with session context – Review agents retrieve the full session history behind a pull request, exposing the tests, attempts, and reasoning that produced the final diff. Reviewers therefore receive grounded explanations rather than inferring intent from the diff alone.
- Evaluations and institutional learning – Sessions are scored against team‑defined rubrics (correctness, test fidelity, surface area, time on task). The resulting findings are promoted into per‑repository guidelines that load automatically for subsequent sessions, turning each run into a learning opportunity.
- Multi‑player sessions – Solo agent runs can be turned into team sessions; any number of participants may join from a browser, driving, contributing, and building together in real time.
- Active session memory – Built‑in MCP tools let humans and agents search the full session archive. When a developer asks, “have we worked on this before?” the agent can surface the exact prior decision point, summarizing the relevant context without reopening closed questions.
The open‑source CLI and the vendor‑neutral event model also mean that teams can switch between Claude Code, Codex, Cursor, or future agents without losing historical data, preserving a single source of truth for all coding activity.
Implications for Enterprise Development Teams
For organizations that have already begun integrating AI coding assistants into CI/CD pipelines, Capacitor addresses a concrete pain point: the coordination overhead that arises when each agent session is treated as an isolated transcript. By preserving context across sessions, the platform makes handoffs instantaneous, PR reviews “grounded” in the actual reasoning that produced a change, and evaluations reusable as institutional knowledge. Kurrent’s VP of Engineering Alexey Zimarev illustrated the impact, noting that his team “was losing critical context every time a session closed,” and that capturing sessions in KurrentDB “changed everything.”
The announcement also ties the product to broader industry dynamics. The 2025 Stack Overflow Developer Survey reports that 84 % of developers are using or planning to use AI tools, with more than half relying on them daily. As AI tools shift from optional productivity boosters to core infrastructure, the need for a shared, searchable memory layer becomes a first‑order engineering management problem. Capacitor’s ability to turn every agent turn into a durable, queryable event means that even short‑term use can make agents “meaningfully better‑performing,” because the system feeds back institutional knowledge into future runs without altering the underlying model.
From a practical standpoint, enterprises can expect several measurable benefits: reduced time spent reconstructing context, faster and more reliable PR reviews, and the ability to enforce per‑repository guidelines automatically. Moreover, because the session archive is searchable by both humans and agents, developers can ask an agent whether a similar problem has been tackled before and receive a concise summary that references the exact prior decision point. This capability transforms the agent from a one‑off code generator into a collaborative teammate that continuously learns from the team’s collective history.
Key Takeaways
- Kurrent Capacitor entered private preview on May 2024, offering free access for Claude Code, Codex and Cursor agents.
- The platform records every agent turn, tool call and reasoning block, storing them as immutable events in KurrentDB’s append‑only stream.
- Six core capabilities—including real‑time collaboration, multi‑agent handoffs, PR‑contextual reviews and institutional learning—are available from launch.
TechInsyte's Take
Capacitor provides a concrete way for enterprises to treat AI‑generated code as a shared, searchable asset rather than a siloed transcript. Its vendor‑neutral event model could simplify multi‑tool strategies, but real‑world impact will depend on adoption rates and how quickly teams integrate the CLI and dashboard into existing CI/CD flows. Buyers should monitor the preview’s performance and any forthcoming pricing or SLA details before committing to production use.
Source: Businesswire
