{"full_name":"OleksandrChekhovskyi/hax","owner":"OleksandrChekhovskyi","name":"hax","shard":"1","url":"https://github.com/OleksandrChekhovskyi/hax","description":"A minimalist, terminal-native coding agent written in C.","homepage":"https://usehax.dev","topics":["ai-agent","anthropic","c","cli","coding-agent","coding-assistant","llama-cpp","llm","local-llm","openai","openrouter","repl","terminal","tui"],"tag_pages":["ai-agent","anthropic","c","cli","coding-agent","coding-assistant","llama-cpp","llm","local-llm","openai","openrouter","repl","terminal","tui"],"language":"C","license":"MIT","avatar":"https://avatars.githubusercontent.com/u/11211653?u=fc03313bb22a2d95c55352eda806b74beff30e96&v=4","created_at":"2026-04-24T12:12:50Z","pushed_at":"2026-11-13T20:14:44Z","metrics":{"stars":1001,"forks":71,"watchers":6,"issues":45,"prs":29,"releases":8,"issues_open":14,"issues_closed":31,"prs_open":8,"prs_closed":20,"prs_merged":1,"discussions":0},"growth":{"day":{"stars":2,"forks":0,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[993,997,999,1001],"forks":[70,70,71,71],"watchers":[6,6,6,6],"issues_open":[14,14,14,14],"prs_open":[8,8,8,8]},"boards":{},"same_owner":[],"issues":[{"number":45,"title":"Qwen 3.8, GLM 5.3 use shell/sed/python to read/write/edit files instead of the tools","comments":9,"created_at":"2026-09-23T10:54:06Z"},{"number":51,"title":"Add a Mistral provider: reasoning is a chunk list in both directions","comments":4,"created_at":"2026-10-08T18:27:32Z"},{"number":50,"title":"Add shortcuts to cycle through reasoning level and favorite models","comments":4,"created_at":"2026-10-03T07:43:16Z"},{"number":16,"title":"Feature request: Typing the next prompt while the current one is in progress","comments":4,"created_at":"2026-08-24T10:43:00Z"},{"number":9,"title":"feat: support direct shell commands","comments":4,"created_at":"2026-08-22T13:45:04Z"},{"number":30,"title":"Allow for colors and glyphs to be customized","comments":3,"created_at":"2026-09-05T12:03:44Z"},{"number":34,"title":"Up arrow always goes to previous prompt, should go up a line first","comments":2,"created_at":"2026-09-09T20:50:38Z"},{"number":26,"title":"feat: session title / alias","comments":2,"created_at":"2026-09-04T10:11:16Z"}],"discussions":[],"latest_release":{"name":"hax v0.7.0","tag":"v0.7.0","published_at":"2026-11-12T09:47:52Z","body":"### Added\n\n- Man pages for offline reference: `man hax` for usage, plus `hax-config(5)`, `hax-sessions(5)`,\n  and `hax-providers(7)`. Packages and source builds install them; the prebuilt binaries do not.\n\n### Changed\n\n- A persistent, live-updating area below the conversation keeps the spinner, elapsed time, and\n  status visible while hax works, even as answers stream and tools run. Pause requests show up\n  immediately, and activity labels change without flickering.\n- Compaction can now recover conversations that already exceed the context limit. Both `/compact`\n  and automatic compaction can summarize a shortened copy, sacrificing older detail first when\n  the full conversation will not fit. With automatic compaction enabled, hax tries to recover from\n  context-overflow errors and continue, even when the server's context size is unknown.\n- Automatic compaction accounts for new tool output and your next message, and leaves room for\n  the summary on small context windows. It runs only when the conversation continues, avoiding\n  unnecessary summaries after a finished reply. Repeated failures suspend automatic compaction\n  instead of retrying indefinitely; one-shot runs now report failures on stderr.\n- Esc lets a compaction whose summary has started finish before pausing; a second Esc cancels it.\n  If compaction ran for a message you just sent, pausing returns that message to the prompt unsent.\n  In one-shot mode, SIGUSR1 likewise lets compaction finish, or pauses immediately if the model\n  has not started answering.\n- Prompt polish: slash-command hints wait until you pause typing, pasted or edited prompts no\n  longer leave empty rows from surrounding whitespace, and quitting removes the empty prompt.\n  Prompt indentation is preserved; whitespace-only input is treated as empty.\n- `/provider` and its completion offer `openai-compatible` and `anthropic-compatible` only once\n  their base URL is configured.\n- Subagent guidance now always explains `--preset <name>`, so the model need not discover how to\n  use a preset you request by name, even when none of your presets has a description.\n- The llama.cpp provider now defaults to port 9931, matching `llama-server` since v0.7.0.\n  For an older server, set `HAX_LLAMACPP_PORT=8080` or `providers.llamacpp.port`.\n","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}