About
Every debugging session, architecture discussion, and code review you have with an AI tool produces valuable data. Today it is scattered across tools, hard to search, and usually locked in vendor formats. RCLM captures it so you can find it, reuse it, move it, and make every next session more efficient.
The problem
Social media platforms built billion-dollar businesses on the content and behavior data their users generated, while users got the product for free and nothing else. We normalized giving away our data in exchange for access.
LLMs are heading down the same path, faster and with higher stakes. The interactions you have with AI tools today are exactly the kind of data that trains the next generation of models. You're contributing to that whether you know it or not.
Your own history — the regex you worked out last month, the architecture discussion that shaped a project, the debugging session you want to show a teammate — is spread across half a dozen tools and formatted in ways only vendors can easily read.
Our premise
RCLM is a capture and observability layer for AI work. It records interactions across providers and tools, then gives individuals and organizations search, usage analytics, compliance controls, and an audit trail.
Sessions you capture belong to you. The immediate value is practical: find old work, understand AI-assisted decisions, continue work in another agent, reduce repeated context, and verify the savings against your own history.
Ownership is the precondition. You cannot choose to keep, move, share, or delete work you do not control. RCLM makes you the owner first.
Capture
CLI-based coding assistants produce the richest structured data, but each capture method works independently and produces the same normalized session record.
The richest capture: messages, tool calls, file diffs.
Native integrations capture structured sessions from Claude Code, Codex CLI, Gemini CLI, Cursor, Antigravity, and opt-in OpenClaw, using the richest events each client exposes.
Zero code changes. Every API call captured.
A local proxy powered by LiteLLM sits between your code and LLM providers. Point API calls at a different address and every request and response is recorded.
Capture ChatGPT, Claude.ai, Gemini — no API key needed.
A Chrome extension captures conversations from ChatGPT, Claude.ai, and Gemini in the background. CLI capture remains the highest-fidelity path.
For individuals
Search and reuse prior work, inspect what an agent actually did, move complete context between tools, reduce redundant tokens, and validate improvements on your own sessions.
Use hybrid semantic and exact search across every captured session, or retrieve prior work directly inside supported AI tools through the RCLM MCP server.
Every session gets a useful title, description, tags, project name, and work category in the background—without a manual filing step.
Review the messages, tool inputs and results, failures, token use, and file changes behind an AI-assisted workflow instead of stopping at its final answer.
See token counts, models, tool usage, failures, categories, projects, and code impact through consistent statistics across capture sources.
Transfer the full captured history between Claude Code and Codex without reducing it to a summary, or export focused context for other assistants.
Range-aware read caching, result shaping, output compaction, deduplication, test filtering, loop detection, and image downscaling reduce avoidable context before it reaches the model.
Read-only MCP replay tools run RCLM's shipped compression rules over your own captured sessions and report token reduction without model calls or task re-execution.
Personal Signals surface repeated restarts, over-exploration, idle gaps, session bloat, repeated discovery, and model mismatch with the evidence behind each pattern.
RCLM Bench provides an open-source CLI and TUI for isolated coding-model comparisons and offline compression replay using Claude Code and Codex sessions.
The Claude Code statusline shows context usage, rate limits, and peak-hour timing while you work, and installs automatically with the native hooks.
Send email-bound, expiring session links to collaborators, then revoke access and review view activity when the work should no longer be available.
Paid users can encrypt full session content with a one-time recovery key while keeping summaries, statistics, and high-level search metadata useful.
Your data, your rules
Nothing is shared, sold, or used for any purpose without an explicit action from you. RCLM does not train on your data.
Export your full history in standard formats, delete individual sessions or your account, and choose region-specific storage when you need it. Credentials and tokens can be flagged or redacted before storage.
Paid users and enterprise organizations can also encrypt raw session details at storage level. Recovery keys are downloaded once, never emailed, and not stored as plaintext. Session summaries, stats, and high-level search metadata stay usable without opening full transcripts.
Enterprise
The enterprise layer is the AI control plane between the tools your developers already use and the oversight your organization needs. It groups users into organizations and teams, tags sessions at ingest, gives admins a cross-developer view, and lets org admins enable encrypted raw session storage without changing the local capture tools each developer already uses.
Developers paste code into ChatGPT, share database schemas with Claude, describe internal infrastructure to Gemini. Some is fine. Some is proprietary source code, API keys, connection strings, or customer data that was never meant to leave your systems. Without visibility, you find out about leaks after the fact — if at all.
LLM usage can balloon fast with coding agents. A misconfigured agent running overnight or a developer sending huge context on every request can create real spend risk, but most teams lack usage breakdowns by developer, team, model, or project.
Which models are your developers using? Are they coding, debugging, writing docs? Are some teams heavy users while others barely touch it? Is AI improving output quality or adding noise? Without data, these are guesses.
For regulated industries or incident response, the inability to answer "what did our developers share with external AI providers over the past 90 days?" is a significant gap. Without a capture layer, that question is simply unanswerable.
What RCLM gives your organization
Current enterprise scope focuses on observability, usage analytics, admin controls, SSO configuration storage, org API keys, encrypted session storage, and retention policy enforcement.
Free for individuals. Each capture method works standalone. No forced pipelines.