albedo
small gleam coding and (not yet) persistent agent daemon with a detachable cli forked from okami.mom/albedo
Star
Fork
Watch

Commits on transcript-readability
cli: theme every screen from terminal colors, add speaker rails and kaomoji
committed
flake: add a dev shell with go, gleam, node and rust
committed
Tell a resumed turn that albedo restarted and mark it in the transcript
committed
Ask before replacing a daemon from another build
committed
List catalog models per profile and isolate malformed profiles
committed
Reject carried proxy state that fails its zlib checksum
committed
Carry provider state through proxy tool-call ids
committed
Carry generation options from requests to every provider
committed
Add an OpenAI-compatible proxy over saved provider profiles
committed
Let extensions serve HTTP from the daemon and pin its port
committed
Tighten the session preview: three lines a message, one gap per exchange
committed
Rebuild the session picker around pins, dates and a transcript preview
committed
Add the Google Antigravity provider
committed
Resolve model providers to upstreams that stream albedo requests
committed
Run OAuth sign-in in the daemon for every client
committed
Cut daemon peak memory to about 80 MB with three large sessions
committed
<system-note> for events sent to model
committed
Add and edit MCP servers through one form
committed
add tg skill
committed
Keep the cached prompt across extension toggles and send only what changed
committed
Capability pages, private MCP credentials, cached-prompt reloads, and global extension defaults
committed
Slim model catalog cache, lock daemon home, and select codex accounts
committed
Detach daemon from launching terminal and fail fast on startup exit
committed
Show compaction shortcut only on compaction events
committed
Fix chat scroll and collapse transcript details
committed
add way to delete providers
committed
remove oauth accounts that 401
committed
mlf
fix(nix): bundle Rebar3 pc plugin for esqlite
committed
readme update
committed
add go tui port with charm
committed
add charm tui skill
committed
view: review the diff first, and fix what a model found using it
committed
readme for render
committed
view: see code as highlighted images for a final review pass
committed
tools can return images to the model
committed
fix tools failing to mention capabilities they had
committed
document the whole model-facing python api
committed
extension pages, starting with /work
committed
work: accept positional arguments for create and list
committed
files: fit the API to how agents actually call it
committed
daemon: type session stages, submissions, and run state
committed
test: strip color codes in the extension picker assertion
committed
fix: let /reload reach the kernel's command catalog
committed
harness: compose each session's extensions from scratch
committed
/reload can accept functions given to it by extensions, ctrl k to view compaction summary
committed
ui changes for queued messages
committed
model switching properly updates default for new sessions + queued message steering
committed
mlf
fix: supervise jobs without leader identity
committed
mlf
test: accept platform-specific temporary paths
committed
WTFPL
committed
retries and ctrl j to view diff
committed
fix read traces
committed
Organize extensions into dedicated directories
committed
Add autoloaded agent instructions extension
committed
/reload models
committed
test/manual: live e2e for the command surface against real deepseek
committed
review-fanout on 9e00cea: typed state seam, prepared-once compositions, naming truth
committed
commands: one catalog drives the CLI menu, user invocation, and typed kernel bindings
committed
trim duplicated wake test coverage
committed
background bash jobs wake idle sessions, local and remote
committed