v0.0.9

Fixed

• The Google provider no longer rejects unknown model ids — any non-empty id is accepted (matching the custom provider), with the known-model table retained for context-window metadata and a 1M-token assumed window for unknown ids. Google retired the gemini-2.5 models on the Developer API with the hardcoded allowlist still pointing at them, leaving every google-provider configuration unusable. Known table now includes gemini-3.5-flash, gemini-3.1-pro-preview, and the floating gemini-flash-latest / gemini-pro-latest aliases.

v0.0.8

Added

• Local markdown ingestion: bo collect ./note.md collects a local .md file as a note — frontmatter stripped (with a warning), title from a leading # H1 when present, content-addressed source URL (bo://note/<hash>) so identical notes dedupe, no LLM summary call.

Changed

• breaking tree layout: leaves now live in leaf/ and branches in branch/ (was root-level .md and branches/). Every domain entity gets its own singular directory. No migration — pre-user break.

Fixed

• Leaf bodies no longer duplicate the title heading when the article's own H1 survives extraction (e.g. a page that leads with a byline before # Title). bo's prepended heading is skipped when the body already opens with a matching H1.
• Code blocks in collected articles are now preserved as fenced markdown instead of being flattened into single-line inline code spans. Multi-line <pre>/<code> blocks keep their line breaks.

v0.0.7

Added

• custom provider — point bo at any OpenAI-compatible endpoint via bo config --provider custom --base-url <url> --model <model>. Any non-empty model id is accepted. Auth via CUSTOM_API_KEY env var or custom_api_key in ~/.bo/auth.json.

Changed

• Compile and collect diagnostics (title-collision warnings, branch-repair notices, pending-recovery notices, per-branch write progress) now print to stderr after the run completes instead of interleaved mid-run. Content and destination unchanged — only the timing shifts.

Fixed

• Tracing/diagnostic output now routes to stderr, keeping stdout clean for command output and --json parsing.

Architecture (non-user-facing)

• The engine is now the public library surface — a vocabulary of capabilities named for what they do, never for the command that invokes them. Query's retrieval, NLP, relevance, context-assembly, and citation-validation layers moved from cli to engine::retrieval with a semantic RetrievalError mapped at the CLI boundary.
• Presentation purity across all workflows: nothing below a command's entry point touches stdout/stderr; workflows return typed results and the CLI renders them. A TUI/REPL/web front-end can now compose the same primitives.
• Compile pipeline stages completed: validation extracted into a dedicated I/O-free trust-boundary module owning CompilePlan, and repair_stale_branches decomposed into a classify → prune → repair → assemble pipeline.

v0.0.6

Added

• zai provider — Z.ai (GLM) API support via the GLM Coding Plan endpoint (https://api.z.ai/api/coding/paas/v4). Includes glm-4.7 (default), glm-4.5-air, glm-5.1, glm-5-turbo, and glm-5.2 models. Auth via ZAI_API_KEY env var or zai_api_key in ~/.bo/auth.json.

Changed

• Untitled leaves now report title: null instead of an empty string in JSON output (bo show, bo status, bo query --json) and in the manifest. Existing manifests load unchanged.

Architecture (non-user-facing)

• Tree/branch/leaf metaphor committed to types — Leaf/Branch domain entities (was LeafRecord/BranchRecord), validated Title/Url newtypes replace bare String aliases.
• Frontmatter serialization unified on typed structs (LeafFrontmatter/BranchFrontmatter) with golden-tested byte-stable .md file output.
• On-disk formats (manifest.json, config.json, .md frontmatter) unchanged — this is an internal type-strengthening, not a format migration.

v0.0.5

Added

• google provider — Gemini API support via bo config --provider google --model gemini-2.5-flash. Includes gemini-2.5-flash-lite, gemini-2.5-flash, and gemini-2.5-pro models with 1M token context windows. Auth via GEMINI_API_KEY env var or google_api_key in ~/.bo/auth.json.
• bo query now retrieves compiled branches in addition to raw leaves — the LLM sees synthesized concept pages alongside source documents and may cite either kind.
• Compile degenerate-result warning: when a Full compile on a large corpus produces a suspiciously thin branch set, bo warns the user without blocking the output. Retry with --all or switch models if the warning appears consistently.
• Architecture documentation: docs/architecture.md (design decisions) and src/AGENTS.md (code rules for contributors / coding agents).
• deepseek-v4-pro model entry (1M context, stronger than v4-flash).

Changed

• Fresh trees auto-run Full compile mode. Incremental mode now only activates when branches already exist, fixing a class of validation errors where the LLM was asked to update non-existent branches.
• bo seed no longer silently defaults to gpt-4.1-mini when no model is given. The model field is required — either via the --model flag or the interactive prompt. bo config without a model now fails with a clear error instead of silently falling back.
• Compile validation hardened: ambiguous leaf title references (two leaves sharing a title) are now rejected rather than silently resolving to the wrong leaf. Leaf slugs are included in the full-compile prompt so models see the canonical identifier.
• Schema generation uses schemars-derived inline schemas (no $ref/definitions/$schema`) — fixes a Google provider rejection and keeps schemas portable across all three providers.
• Type-level schema enforcement: the LlmProvider contract now requires normalized schemas at compile time, not runtime.
• Compile output (--json) no longer reports a redundant context_mode field (it was always 1:1 with mode).
• Three different LLM runtime patterns unified behind a shared OnceLock-backed runtime. All CLI modules share one long-lived async executor rather than building fresh ones per call.

Fixed

• Branch .md frontmatter now correctly drops deleted leaf filenames from the leaves: list when a leaf is removed. Previously the manifest was updated but the branch file on disk was stale.
• Google provider compile returning HTTP 400 on schema fields injected by schemars — fixed by inlining subschemas.
• Duplicate-URL detection on single-URL bo collect — a regression from an internal refactor that dropped the pre-fetch duplicate check.
• Five distinct compile failure modes are now caught by the validation gate and refuse to write, rather than silently producing a corrupted or degraded tree.

Architecture (non-user-facing)

• Compile pipeline refactored: mode-delta logic split into focused functions, duplicated preflight collapsed, CompileContextMode redundant type removed. God-function eliminated; pipeline reads as a traceable progression.
• Retrieval scoring deduplicated: leaf and branch scoring share a single parameterized function.
• Atomic manifest-commit sequence extracted from duplicate collect/compile sites into a shared engine::pending helper.
• pub(super) visibility discipline enforced throughout compile and query modules — no internal plumbing leaks across commands.

v0.0.4

Fixed

• bo collect — LLM summary generation no longer falls back to the deterministic summary on OpenAI providers. The structured-output schema was being rejected by OpenAI's strict-mode validator because summary was missing from the schema's required list. DeepSeek users were unaffected.

v0.0.3

Fixed

• npm install — bo no longer fails with use strict: command not found / syntax error near unexpected token '(' after npm install -g @skillicinski/bo. The wrapper script was missing its #!/usr/bin/env node shebang, causing the OS to execute it as a shell script.

v0.0.2

Added

• npm distribution — install with npm install -g @skillicinski/bo. No Rust toolchain needed. Pre-built binaries for macOS (Intel + Apple Silicon) and Linux x86_64.
• DeepSeek provider support (deepseek-v4-flash, deepseek-v4-pro). Select with bo config --provider deepseek.
• bo status command — shows tree health and compile readiness.
• bo compile --all flag — recompile the full corpus and allow complete branch graph rewrite.
• bo collect accepts multiple URLs in one invocation and .txt files containing one URL per line.
• compile_model config key — pin a heavier model just for bo compile. Set with bo config --compile-model <id>.
• Compile validation gate — bo compile rejects responses that would result in no file changes and surfaces an actionable error.
• Interactive confirmation gate on bo raze — operators must type yes to proceed. No --force / --yes override (by design, so agents cannot bypass it).
• manifest.json is now the canonical record of tree topology and metadata. Lives at {tree}/.bo/manifest.json. Written atomically (tmp + rename) on every mutation.

Changed

• bo compile is incremental by default — only processes leaves collected since the last compile. Use --all for full recompile.
• bo config is now flag-driven: bo config --provider <name> --model <id> --compile-model <id>. Replaces the previous bo config auth / bo config get / bo config set subcommands.
• Default model changed from gpt-4o to gpt-4.1-mini.
• bo show now defaults to a frontmatter card view; use --full for the complete body.
• bo list reworked into a branch-centric tree view by default. New flags: --branch <name>, --recent, --limit <n>.
• All read commands (bo status, bo list, bo show, bo query, plus bo collect duplicate detection) consult the manifest. They no longer read index.jsonl, state.json, or scan the branches/ directory for metadata.
• bo collect and bo compile write the manifest as the primary commit, then mirror to index.jsonl / state.json / branch frontmatter as a transient safety net (removed in v0.0.2 stage 3b).
• Branch .md frontmatter renamed compiled_at → created_at for clarity. updated_at unchanged. The manifest's BranchRecord mirrors these names.
• bo raze now preserves stored provider credentials by default; use --include-auth for a full credential wipe.
• bo status no longer counts leaves with unparseable frontmatter as "uncompiled". Such leaves were always also flagged as skipped by health output; the duplicate signal is gone, the explicit skipped signal stays.
• Compile context-overflow and validation errors now reference the new flag-based config (bo config --compile-model <id> / bo config --model <id>).

Removed

• bo search command. Use bo list --terms <terms> for title/slug filtering.
• bo config auth subcommand. Save API keys via the OPENAI_API_KEY / DEEPSEEK_API_KEY env vars or by hand-editing ~/.bo/auth.json with flat keys (openai_api_key, deepseek_api_key).
• bo config get and bo config set subcommands. Replaced by the flag-driven bo config.

Recovery

• If manifest.json is missing on read, bo reconstructs it from the secondary store (index.jsonl + branch frontmatter) on the fly, persists the result, and prints manifest missing; reconstructed from secondary store to stderr. This affordance is removed in v0.0.2 stage 3b along with the secondary store itself.

v0.0.1

First experimental release.

Added

• bo seed, bo collect, bo compile, bo query, bo list, bo search, bo show, bo raze
• bo config auth --provider openai — store API key locally
• bo config set model / bo config get model
• --json flag on all commands
• YouTube transcript collection
• Zero-citation detection (refuses hallucinated answers)
• Install smoke test in CI

Notes

• OpenAI-compatible providers only
• Lexical retrieval (no embeddings)
• Requires Rust toolchain to install