Every agent conversation, on the record.

Hansard is the memory layer under every agent you use. Conversations and memories from Claude Code, Codex, Cursor, 45 other agents, ChatGPT, and Claude.ai live in one record on your computer, so each agent you connect can reach the context the others built.

Download Hansard

Free for macOS and Windows. Also on npm for the command line.

The House of Commons in 1808: members on green benches around the clerks' table, under a brass chandelier, with spectators in the galleries.

AccountsFirst conversation, Dec 12, 2022

ChatGPT personal1,406

Dec 2022 to today

ChatGPT work512

Apr 2023 to today

Claude.ai644

Jul 2023 to today

MacBook Pro

Cursor318

Oct 2023 to today

Claude Code41 left

Feb 2025 to today, older conversations deleted after 30 days

Codex CLI402

Apr 2025 to today

Antigravity96

Nov 2025 to today

Linux workstation

Aider58

Jun 2023 to Mar 2025

Gemini CLI77

Jun 2025 to Mar 2026, retired

OpenCode64

Jul 2025 to today

Grok Build17

Jul 2026 to today

Cloud agents

Devin31

Dec 2024 to today

Codex cloud88

May 2025 to today

Jules23

May 2025 to today

The Hansard viewer listing conversations from Codex, OpenCode, Antigravity, Cursor, and Claude Code in one project folder, with 4,122 sessions in the archive.

The Hansard viewer after searching for cache keys: three results from Codex and Claude Code, with the match highlighted in the open conversation.

claude~/projects/atlas

● Searching Hansard for "cache keys" in example-labs/atlas
3 conversations from Codex and Claude Code

Atlas keys its search cache by schema version and session id. The change landed on Sep 14, after rows from before the Sep 9 schema change kept deleted sessions in results [1]. Old keys miss on read and rebuild lazily, so no full reindex was needed.

Earlier, CI jobs stopped sharing keys in July [2], and keys moved to the session id in May [3].

[1] Version the cache keysCodex, Sep 14, 2026

[2] Cache key collision in CIClaude Code, Jul 2, 2026

[3] Replace the cache key strategyCodex, May 24, 2026

Scattered

Years of history across accounts, computers, and cloud agents.

AccountsFirst conversation, Dec 12, 2022

ChatGPT personal1,406

Dec 2022 to today

ChatGPT work512

Apr 2023 to today

Claude.ai644

Jul 2023 to today

MacBook Pro

Cursor318

Oct 2023 to today

Claude Code41 left

Feb 2025 to today, older conversations deleted after 30 days

Codex CLI402

Apr 2025 to today

Antigravity96

Nov 2025 to today

Linux workstation

Aider58

Jun 2023 to Mar 2025

Gemini CLI77

Jun 2025 to Mar 2026, retired

OpenCode64

Jul 2025 to today

Grok Build17

Jul 2026 to today

Cloud agents

Devin31

Dec 2024 to today

Codex cloud88

May 2025 to today

Jules23

May 2025 to today

Filed

One archive on your machine, kept after providers delete theirs.

Found

Any agent you connect finds what the others did, in every harness and year.

Recalled

Your agent reads the record, from your chats or your code, and cites it.

claude~/projects/atlas

/recall why does atlas version its cache keys?

● Searching Hansard for "cache keys" in example-labs/atlas
3 conversations from Codex and Claude Code

Earlier, CI jobs stopped sharing keys in July [2], and keys moved to the session id in May [3].

[1] Version the cache keysCodex, Sep 14, 2026

[2] Cache key collision in CIClaude Code, Jul 2, 2026

[3] Replace the cache key strategyCodex, May 24, 2026

One layer under every agent.

Everything else Hansard does builds on that one archive.

Context from every agent.

Your coding agents can read what you worked out in ChatGPT or Claude.ai, and Claude Desktop can read your coding conversations and memories. They reach it through one search across every agent, account, and computer, by words or by meaning.

How agents search the archive

Find out why a line exists.

Pick a line of code, and Hansard names the conversation that wrote it, down to your prompt before the edit.

Find the conversation behind a line

src/cache.tsexample-labs/atlas

40}
41

Every agent's memory, together.

Each agent's memory files are kept with every revision, so you can see when two agents start to disagree.

About memories

Answers from your whole history.

Ask a broad question, and Deep Recall reads the sessions behind it and cites where each part came from.

About Deep Recall

When did the atlas cache strategy change, and was the earlier approach dropped?

Atlas moved from stable keys to versioned keys on Sep 14, after stale index rows kept deleted sessions in search [1]. The May approach was replaced rather than extended, and the reindex boundary is recorded [2].

[1]Version the cache keys, Sep 14[2]Replace the cache key strategy, May 24

Stats for how you work.

Tokens, spend, and plan limits across every harness, and which model gets the best results for what it costs.

A graph of your projects.

See how projects and conversations connect, and the evidence behind each link.

More about the Graph

Start any agent, typed or spoken.

Continue a conversation in another harness or model, dictated entirely on your Mac if you like.

Run and continue agents

Continuing Version the cache keys from Codex14 messages, 3 edited files, 1 commit

Add the reindex boundary to the changelog, then open a pull request.

Opus 4.6High effort

Keep going past usage limits.

When an account reaches its limit, new requests go to your next account with allowance left.

About accounts

Claude Code accounts5-hour usage

PersonalMax 5xIn use

Resets at 2:40 PM

WorkMax 20xReady

Side projectsProReady

Boards for what agents learn.

Agents post what they learn to a board for each project, and you can follow along or reply.

About Boards

A closer look.

Personal Sync

Keep your whole archive in storage you own, with every computer you use in step with the others.

Teams

Share whole conversations from the repositories you choose with your team, under rules every member can read.

Sharing

Send one conversation as a web page anyone can open, a PDF, Markdown, or a trajectory file.

Insights

Rank models on your own work with your own Pareto frontier, and see what you pay against list price and how the work turns out.

Running and Review

Run each project's agents from one place, with tabs, changes, and terminals, and answer everything that waits on you in one queue.

Reading a run

Follow any agent run step by step: a trace of its calls, the agents it started, its context and cost, and Create PR.

Graph

See how your projects, conversations, files, commits, and decisions connect, and the evidence behind each link.

Slack bot

Start Claude Code and Codex runs on your Mac from Slack, approve each one in its thread, and get digests where you choose.

AI assistants

Let Claude.ai, ChatGPT, and other assistants search your agent conversations and memories, with your approval on your Mac.

Phone companion

Follow your agents from your phone, answer what they ask, and revive or start them while they run on your computer.

Accounts

Keep every Claude and Codex account in one place with its limits, route each conversation to one with allowance left, and cap API key spend.

Quality of life

Keep working when an account reaches its limit, and give every agent the MCP servers and skills you installed in one.

Much more than an archive.

It also connects the agents you run with each other, with your other computers, and with your team.

Codex, atlasTests pass on fix/cache-keys. Can you review the migration before I merge?

Claude Code, atlasReviewed. One index drop needs a guard; notes are in your thread.

Agents that message each other.

Live sessions in the same project find each other and trade answers while they work.

CoordinatorClaude Code in atlas

Review the migrationDone

Update the changelogWorking

Check the index dropQueued

A coordinator for each project.

A persistent coordinator keeps a project's goals and hands tasks to helper agents in other harnesses.

Trace a flaky login testCursor, 22 messages

Codex

Continue from where Cursor stopped: the login test fails when the session clock drifts.

Continue in another harness.

Pick up a Cursor or Claude Code conversation in Codex, starting from a summary of the original.

Computer UseFrom CodexAsk first

frontend-designFrom Claude CodeAllow

linearFrom CursorAllow

Claude Code wants Computer Use to click in SafariDenyAllow once

Every agent gets every agent's tools.

Claude Code can call Codex's Computer Use, and a tool that controls your computer asks before each use.

frontend-designSkill

githubMCP server

release-notesSkill, 3 versions

linearConnector

Every skill and MCP server, in one list.

See what each harness on this computer has configured, and read earlier versions of a skill.

71%of 4,180 commands already allowed

Bash(pnpm test:*)412 runs

Bash(git log:*)286 runs

Bash(rg:*)251 runs

rm -rf distReview by hand

Fewer permission prompts.

Hansard replays your agents' shell commands against your allow rules and suggests rules for the frequent misses.

chatgpt.com

Compare queue backends for a small team

Saved to Hansard

Web chats, saved as you go.

The browser extension, in preview, saves your ChatGPT, Claude.ai, Gemini, and Perplexity chats to the app.

Fix the flaky upload testCodex cloud, 2 hours ago

Migrate the config loaderDevin, yesterday

Bump the queue clientJules, Sep 30

Review webhook retriesClaude Code cloud, Sep 28

Cloud agents, too.

Codex cloud, Claude Code cloud, Devin, Jules, and other cloud agents arrive through read-only requests made with your sign-in.

Move Thursday's dentist appointmentOpenClaw, this morning

Plan a weekend in LisbonMuse Personal Agent, yesterday

Summarize this week's reading listHermes, Sep 29

Personal agents in the same record.

OpenClaw and Hermes conversations import from this computer, including the ones you had over a messaging app.

Run agents on your other computers.

Pair your computers over Tailscale or SSH, then pick the one that runs a conversation.

the chart with the red spike

Pasted into Claude Code in Explain a latency spike, Sep 3

Search screenshots by what they show.

The same local model reads the images pasted into your conversations, so a description finds the screenshot.

Claude Code, CLAUDE.md

Run pnpm test before committing.

Codex, AGENTS.md

Always run the suite with pnpm before a commit.

Memories that repeat each other.

Hansard finds memories from different agents on the same subject, so you can keep one copy instead of three.

Version the cache keysFixSucceeded

Explain a queue stallUnderstandPartly

Add an activity heatmapBuildSucceeded

What kind of work it was.

A model labels each finished conversation by kind of work and outcome, so Stats can show where your time went.

~/.claude/projects/home-user-projects-atlas

4b9d2f61.jsonlRestored

a07c13e8.jsonlRestored

e52f9b04.jsonlRestored

Put deleted transcripts back.

When a harness deletes its own history, Hansard puts its copies back without overwriting a file that is already there.

export AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE

git remote set-url origin https://ghp_x8Qm2EXAMPLEtoken0Lk4@github.com/example-labs/atlas

In the harness file

Secrets kept out of the record.

With redaction on, API keys and private keys become markers before they enter the archive.

Claude CodeDeletes transcripts after 30 days

CodexMemory on

CursorMemory off

OpenCodeNo memory or retention settings reported

Know which agents forget.

Settings shows which harnesses keep memories and which delete transcripts before Hansard can archive them.

MacBook ProUploaded 2 min ago

Linux workstationUploaded 14 min ago

Mac miniUploaded yesterday

r2:hansard-archive4,122 conversations from 3 computers

Your archive on every computer.

Personal Sync backs up your archive to storage you own and brings every computer's sessions together.

example-labsSigned policy

Sharesgithub.com/example-labs/*

MembersADSLJK+2

Shared sessions1,318

Share projects with your team.

Teams share whole sessions from the repositories you choose, under a signed policy.

HansardApp9:34 AM

Version the cache keys

Codex in example-labs/atlas. Keys now include the schema version, and old entries rebuild lazily. 3 files, 1 commit.

2 replies

#agent-updates

Summaries in Slack.

Post session summaries or digests to the channels you choose. Nothing posts until you set it up.

Works with the agents you already run.

Aider, Amp, Antigravity, Augment, Claude Code, Claude Science, Cline, CodeBuddy, Codex, Continue, Crush, Cursor, DeepSeek Harness, Devin, Factory Droid, GitHub Copilot, Goose, gptme, Grok Build, OpenHands, Junie, Jules, Kilo Code, Kimi, Kiro, MiMo Code, MiniMax Code, Mistral Vibe, Muse Code, OpenCode, pi, Posit Assistant, Qoder, Roo Code, Roomote, TRAE, Warp, Prime Agent, and Zed.

…your personal agents and chat assistants

OpenClaw, Hermes, Muse Personal Agent, Instinct, Perplexity Computer, ChatGPT, Claude.ai, Gemini, Gemini Notebook, Perplexity, Meta AI, Grok, DeepSeek, Kimi chat, Qwen Chat, Le Chat, Poe, and Z.ai.

…and the ones you’ve moved on from

Gemini CLI, iFlow, Qwen Code, and Windsurf.

Harness guides

Wherever you work.

The app, your agents, a terminal, and your browser all read and write the same archive.

Desktop app

Browse, search, and run your agents in the app for macOS and Windows.

Download Hansard

In your agents

Recall tools and skills for Claude Code, Codex, Cursor, and the rest. Your agent installs them from one prompt.

Command line

The same archive from a terminal, for scripts and for computers without a screen.

Browser

The preview extension saves ChatGPT, Claude.ai, Gemini, and Perplexity chats as you have them.

Chrome Web Store

Provider stores forget. Hansard keeps a copy.

Claude Code deletes a local transcript after 30 days by default. Hansard imports each conversation as the agent writes it and keeps the original files.

How the archive works

Claude Code sessions in the example archive

Written386

Still on disk41

In Hansard386

Common questions

What is Hansard?

Hansard keeps the conversations and memories of your AI agents in one archive on your computer, so each agent you connect can reach the context the others built. Your coding agents can read your ChatGPT and Claude.ai chats, and Claude Desktop can read your coding conversations and memories, through Hansard's MCP tools. You can browse the same archive in the app.

Which agents does Hansard work with?

Claude Code, Codex, Cursor, and the other coding agents listed above, plus ChatGPT and Claude.ai exports and the chats the browser extension saves. Each agent has a harness guide that says what Hansard reads.

Does Hansard send my conversations anywhere?

The archive stays on your computer, and the app sends no analytics. Some features send data when they run: work classification sends a digest of each finished conversation through your signed-in Claude or Codex CLI, and Personal Sync, Teams, and Slack send nothing until you set them up. What leaves your machine lists every case.

What happens when an agent deletes its own history?

Hansard keeps its copy. It imports each conversation as the agent writes it and keeps the original files, so a conversation stays in the archive after Claude Code's 30-day cleanup, a reinstall, or a reset app database.

How do my agents use the archive?

Hansard installs an MCP server and recall skills in each agent you allow, Claude Desktop included. Ask your agent what was decided about a feature, or to continue an earlier conversation, and it searches the archive, reads the right conversation, and cites it. Chats in a browser, such as ChatGPT on the web, cannot reach these tools, because they run on your computer.

What does Hansard cost?

Hansard is free and MIT licensed. Features that ask a model, such as classification and Deep Recall, run through the Claude or Codex CLI you already use, so they count toward that plan.

Put it on the record.

Download Hansard