v0.2.1

What's Changed

• Let a tag reach the registries it has never once been able to reach by @asanabrial in #93

Full Changelog: v0.2.0...v0.2.1

v0.2.0

What's Changed

• Let a tag reach every registry, instead of four acts and one memory by @asanabrial in #30
• Make publishing something you can ask for, not only something a tag does by @asanabrial in #31
• Write down where cargo binstall finds the binary, rather than relying on a guess by @asanabrial in #32
• Make the README's GIF a recipe rather than a binary nobody can regenerate by @asanabrial in #33
• Say that Homebrew and Scoop exist, since they now do by @asanabrial in #35
• Put the four Docker files in a directory, so the README starts three rows sooner by @asanabrial in #36
• Put the four install scripts in scripts/, while nobody has copied the URL yet by @asanabrial in #37
• Let the registry job ask first, the way the other two already do by @asanabrial in #38
• Record the loop, without the browser that stopped working by @asanabrial in #39
• Show both recordings, each beside the sentence it is of by @asanabrial in #40
• Correct five things the README claimed and one it contradicted by @asanabrial in #41
• Run the unit tests on the three platforms a runner can execute by @asanabrial in #44
• Test arm64 Linux on the runner that already builds it by @asanabrial in #46
• Let the jobs that catch the worst regressions block a merge by @asanabrial in #49
• Stop this file asserting counts it does not own, and dates it cannot show by @asanabrial in #50
• Check the counts these comments state, instead of correcting them again by @asanabrial in #52
• Count the set the sentence describes, and stop the docstring counting at all by @asanabrial in #54
• Publish the image for the architecture most of its users are on by @asanabrial in #55
• Hold the two tag pattern lists to each other, since nothing else does by @asanabrial in #61
• Cut the title in the two sections that print a preview, like every other line by @asanabrial in #64
• Answer in public the question this project was launched asking by @asanabrial in #66
• Put the largest cost back on our own ledger, and say which sample the headline is by @asanabrial in #68
• Draw the result that took thirteen runs, instead of writing it in a sentence by @asanabrial in #71
• Make leteo compatible with ZCode and DeepSeek Harness by @asanabrial in #72
• Give ZCode a plugin bundle, and stop the bundle guards watching only two by @asanabrial in #73
• Correct an install instruction for a command ZCode does not have by @asanabrial in #74
• Stop writing a patch file the harness cannot parse, on the shape it ships by @asanabrial in #75
• Publish the tool list in the revision the server accepts by @asanabrial in #77
• Strip the comments that narrate, keep the ones that constrain by @asanabrial in #78
• Stop testing comments, and drop the counts that needed the testing by @asanabrial in #80
• Count six months one way, and repair the clocks counted the other way by @asanabrial in #83
• Build the arm64 half of the container image on the pull request, not at the tag by @asanabrial in #81
• Say what the package-link probe measured, in place of the untested sentence by @asanabrial in #84
• Hold the premise the leader check rests on, instead of stating it by @asanabrial in #86
• Point CI's container cache at the scopes the release reads, and hold them there by @asanabrial in #88
• Name this release in the changelog, and write the four fixes that were missing from it by @asanabrial in #91

Full Changelog: v0.1.2...v0.2.0

v0.1.2

What's Changed

• Take the scope npm insists on, and say why in the package's own README by @asanabrial in #25
• Download with node:https, because fetch does not arrive by @asanabrial in #26
• Refuse to write a path npm owns, instead of breaking quietly later by @asanabrial in #28
• Release 0.1.2, so that npx actually downloads by @asanabrial in #29

Full Changelog: v0.1.1...v0.1.2

v0.1.1

What's Changed

• Explain the README's commands, and move the guards that never read it by @asanabrial in #1
• Fix the wordmark's mismatched letters, and show the dashboard in the README by @asanabrial in #2
• Lead with what using it is like, in half the words by @asanabrial in #3
• Read the recall floor from one place, so measuring it measures the product by @asanabrial in #4
• Give the server a name the registry can check, and say why the line is visible by @asanabrial in #5
• Write down that the third stage's floor cannot be tuned, and why nobody should try by @asanabrial in #6
• Point the README at the write-ups, since a document nothing links is not published by @asanabrial in #7
• Give the MCP server its own image, and make CI start it rather than just build it by @asanabrial in #8
• Say who maintains this, so the directory listing can be claimed by @asanabrial in #9
• Say in mem_save itself that candidates are work, not information by @asanabrial in #10
• Point the install-from-source line at the crate, now that there is one by @asanabrial in #11
• Move four files off the root, and leave the ones a published URL points at by @asanabrial in #12
• Arrange the contention this test needs instead of fighting for it by @asanabrial in #22
• Pin the Linux builds to a glibc floor somebody chose by @asanabrial in #21
• Say that the plugin is there, after shipping two of them in silence by @asanabrial in #16
• Hold the plugin bundles to their matchers too, and find Codex already drifted by @asanabrial in #17
• Add an npm wrapper so npx leteo works without a toolchain by @asanabrial in #23
• Give the two plugin bundles the README their layout says they have by @asanabrial in #20
• Bind the six versions to one, and set them to 0.1.1 by @asanabrial in #24

New Contributors

• @asanabrial made their first contribution in #1

Full Changelog: https://github.com/asanabrial/leteo/commits/v0.1.1

v0.1.0

First release of Leteo — persistent memory for AI coding agents. An MCP server
and a set of lifecycle hooks over one local SQLite database, so the agent keeps
what was decided, fixed and learned across sessions and compactions without you
ever prompting it to remember.

What's in it

• A 22-tool MCP server over stdio, and lifecycle hooks for session start,
  compaction recovery, prompt capture, subagent capture and session stop.
• Setup for twelve clients: Claude Code, Codex, Cursor, Gemini CLI, OpenCode,
  Windsurf, VS Code Copilot, Kilo Code, Qwen, Kiro, Antigravity and Pi.
• Full-text search with weighted BM25 and a widening retry, so a question that
  matches none of its words strictly still finds the memory.
• Memories with types, topic keys, a review clock, and a relation graph that
  says which memory a later one overturned.
• A JSON command line and an interactive terminal UI.
• Optional PostgreSQL cloud replication, per project, off by default.
• leteo import --from-engram reads an existing Engram database directly.

Twelve languages for the interface and for what memories are written in.

Binaries for x86-64 Linux, Windows and macOS, and arm64 Linux and macOS, each
verified against SHA256SUMS.

The complete notes, including what is measured, what is deliberately missing and
the known limits, are in CHANGELOG.md.

Full Changelog: https://github.com/asanabrial/leteo/commits/v0.1.0