This is a public copy of a conversation between Tiles and 
prepare release notes for last 2 versions
Reasoning details
We need to prepare release notes for last 2 versions. We must check the project repository structure. Let's explore.
ls -R .
10
We need to prepare release notes for last 2 versions. We need to know what the project is, likely a library called "tiles". Release notes probably refer to changes between last two releases (tags). We need to inspect git history. Let's see git log.
git log --oneline --decorate --pretty=format:%h %s
10
We need to prepare release notes for last 2 versions. We need to know the commit history and the changes in those commits.
We cannot see entire repository content easily but we can run git log. Let's try simple:
git log --oneline --decorate
10
The user wants "prepare release notes for last 2 versions". Let's parse: They want release notes presumably in some format. The repository appears to include a "repl.rs" maybe for releases? Search for release notes file.
./CHANGELOG.md
The user wants: "prepare release notes for last 2 versions". They likely want a changelog or release notes summarizing the changes in the most recent two releases. According to CHANGELOG.md, we have entries for 0.4.13 and 0.4.12. The last two versions are 0.4.13 and 0.4.12. So we need to prepare release notes for those two versions.
We should output a concise release note content. Use the format from CHANGELOG.md? The change log entries include "Added", "Changed", "Fixed". We should compile release notes summarizing the changes in each version.
We can output:
For 0.4.13 (2026-06-19):
- Added flags for llama.cpp tuning in tiles run: --context-length, --gpu-layers, --offload-kqv, and --batch-size
- Added daemon /config endpoint for python inference backend to read config
- Persist llama.cpp settings under [llama] in config.toml instead of env variables
- Reload Linux llama.cpp runner on config change even if same model path
- Reworked link command UX for tiles link create, add, list-peers, revoke; support offline link codes and UCAN tokens
For 0.4.12 (2026-06-15):
- Added full support for Linux: model inference, keychain mgmt, installer
- Using UCAN tokens for authorization in p2p syncing; added subcommands create-token and add-token
- Changed tiles server to inference with run-background flag
- Added metadata about tool calls in session records
But we should include maybe the main new features.
Thus produce release notes accordingly.
Release Notes – 0.4.13 (2026‑06‑19)
New features & improvements
| Feature | Description |
|---|---|
| Llama.cpp configuration | Persist llama‑cpp settings under the [llama] section in config.toml. The daemon /config endpoint now lets the Python inference backend read this config. |
| Command‑line tuning | The tiles run command now accepts the following flags for fine‑tuning llama‑cpp: --context-length, --gpu-layers, --offload-kqv and --batch-size. |
| Link‑command UX | The interactive link commands (tiles link create, add, list-peers and revoke) have been refactored for a cleaner UX. They now natively support both offline link‑codes and UCAN tokens. |
| Daemon endpoint | A new /config daemon endpoint provides the Python inference backend with easy configuration access. |
Release Notes – 0.4.12 (2026‑06‑15)
New features & enhancements
| Feature | Description |
|---|---|
| Full Linux support | Full support for model inference (llama‑cpp), key‑chain management and the installer. |
| UCAN‑based auth | P2P syncing now uses UCAN tokens for authorization. New sub‑commands tiles link create-token <DID> and tiles link add-token <token> enable token creation/registration. |
| Inference command rename | The tiles server command was renamed to tiles inference. It now includes a run‑background <bool> sub‑command that keeps the Tiles REPL alive when set to true. |
| Tool‑call metadata | Tool calls now record richer metadata. Session records contain tool name and arguments for better debugging & audit. |
These notes summarize the key changes introduced in the last two releases of this project.
