Context window
- Requests asked for the model's full max_output, but providers count
input + output against the window: long chats with large-output models
(DeepSeek V4.1 Flash, 384k) failed with "input too long" before automatic
compaction (85 %) kicked in. max_tokens is now what the window leaves
(estimated input = last usage + messages since), the chat is compacted when
less than 32k would remain for an answer, and a context-overflow error
compacts and retries once.
- ModelConfig.max_input (catalogue limit.input, editable in the provider
settings) lowers the compaction threshold for models with an input limit.
Enforced reviews (modes with refine.rounds > 0, refine.after_todos default on)
- Every todo marked completed is reviewed against the work since the last
review; with findings it is reopened and the feedback goes to the agent
(max. rounds per todo). Passed reviews leave a visible note.
- Ending the turn with open todos sends the agent back to work while it makes
progress (max. 10 reminders in a row without progress); such modes may take
up to 2,000 steps per turn. Reaching the step limit is reported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Fetch prices for opencode (Zen and Go) incl. limits, image input, reasoning,
long-context tiers and peak-hour/date factors; costs apply them per request
- Record the cost opencode reports per response; track Go allowance in 5h/7d/month
windows against a provider plan limit (budget table, status overlay)
- Settings → Config file: move values of ${NAME} references into the config
- Examples, presets and forms keep keys in the config instead of .env
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- codecraft: full-screen terminal client (ratatui) – local tools and files,
models via the CodeCraft server gateway or own providers (standalone with a
guided setup), slash commands, @files, permission dialogs, resume, update
- Static Linux CLI builds (x86_64/aarch64, cargo-zigbuild) served by the
server: /cli/install.sh and /cli/download/<target>
- Project memory (remember tool, .codecraft/memory.md) and personal memory
- Model fallback when a host is unreachable or a budget is used up
- Version/commit in UI and health, daily release check, "What's new" dialog
- PWA: install button, maskable icon, manifest id and screenshots
- Case-insensitive detection of connection errors for retries
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Personal API tokens (full/read scope, expiry, revoke) and CLI password login
- Cargo workspace: codecraft-core (types, providers, permission rules) shared by
server and the upcoming terminal client; server re-exports the old paths
- CLI server API: /cli/config, /cli/skill, /cli/tool (server MCP tools),
/llm/stream model gateway with budgets and usage per user
- Theme/language preferences mirrored to the server; personal memory endpoint
- Statistics: provider and mode tables stacked; CLI usage labelled
- cli/: work in progress (terminal client is being rewritten as a local TUI)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Richard ZinkandClaude Opus 5.5
Richard ZinkandClaude Opus 5.5