Heads up This site is currently under heavy development.
Subscribe Get it delivered — the daily firehose, filtered to the tools you run, plus the documentation changes vendors never announce. Compare plans →

The AI Toolchain — issue -156, March 16, 2026

THE AI TOOLCHAIN NO. -156
Tail
THE DAILY RELEASE FIREHOSE
PUBLISHED MARCH 16, 2026 · EVERY WEEKDAY
EDITIONS tail grep head diff uniq

The daily firehose — everything the toolchain shipped today, already filtered.

// HOW THIS ISSUE IS MADE

We read every release from the 174 tools on our watchlist at the source — GitHub and GitLab release notes, vendor release pages and changelogs, project blogs and feeds, vendor press releases, and the source code behind the tag. Bug-fix-only releases and non-product newsroom noise are dropped; what's left is summarized down to the new capability, how to try it, and any screenshots or videos the release itself published. Every entry links to the sources it was built from.

VIEW
ISSUE VIEW full issue
Do you prefer this view?
$ tct list   # 8 tools matched
AI & LLM Tooling
◆  AI Coding Agents

Cline

Sources Release notes → v3.73.0 NOTES

Autonomous coding agent as an SDK, IDE extension, or CLI assistant.

Cline v3.73.0 adds W&B Inference by CoreWeave (17 models) and improved parallel tool calling for OpenRouter and Cline providers.

└──▷ GET THIS VERSION
$ git clone --branch v3.73.0 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout v3.73.0
  • Adds W&B Inference by CoreWeave as a new API provider with 17 available models.
  • Improves parallel tool calling support for OpenRouter and Cline providers.
Was this useful?

Charm Crush

Sources Release notes → v0.50.0 NOTES

Glamourous agentic coding for all

crush run gains --continue and --session flags to resume past sessions non-interactively

└──▷ GET THIS VERSION
$ git clone --branch v0.50.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.50.0
└──▷ TRY IT
Pick up where you left off in the most recent session without opening the interactive UI.
$ crush run --continue "tell me more"
Continue a specific past session by ID — useful in scripts where you track session IDs from crush session list --json.
$ crush run --session <session-id> "um, can you repeat that?"
  • Adds --continue (-C) flag to crush run to resume the most recent session non-interactively.
  • Adds --session (-s) flag to crush run to resume a specific session by ID non-interactively.
Was this useful?

GitHub Copilot CLI

Sources Release notes → v1.0.6 NOTES

Copilot CLI v1.0.6 adds dynamic tool search for Claude models, cross-platform hook configs, Open Plugin spec support, and expanded GitHub MCP tool flags.

└──▷ GET THIS VERSION
$ git clone --branch v1.0.6 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.6
└──▷ TRY IT
Load custom instruction files from a project-specific directory so the agent picks up team guidelines automatically.
$ COPILOT_CUSTOM_INSTRUCTIONS_DIRS=./team-instructions copilot
Load a plugin from a local directory so .claude-plugin/plugin.json is discovered and the plugin's tools are available to the agent.
$ copilot --plugin-dir ./my-plugin
  • Adds --enable-all-github-mcp-tools, --add-github-mcp-toolset, and --add-github-mcp-tool flags now taking effect in SDK (ACP mode).
  • Adds skipPermission per-tool setting so extension tools can bypass permission prompts individually while still using the permissions system.
  • Adds COPILOT_CUSTOM_INSTRUCTIONS_DIRS environment variable support for loading custom instruction file paths correctly.
  • Hook configuration files now accept PascalCase event names alongside camelCase, enabling a single hook config to work across VS Code, Claude Code, and the CLI without modification.
  • Hook config files now support Claude Code's nested matcher/hooks structure and optional type field.
+9 moreshow less
  • Adds support for the Open Plugins spec: .lsp.json file locations, exclusive path mode, and : namespace separator for loading plugin and marketplace manifests.
  • Plugins using .claude-plugin/plugin.json are discovered when loaded via --plugin-dir.
  • Instruction file frontmatter applyTo field now accepts both string and array values.
  • Enables dynamic tool discovery and use with tool search for Claude models.
  • Sub-agents launched by the task tool receive human-readable IDs based on their name (e.g., math-helper-0) instead of generic agent-0 identifiers.
  • The create_pull_request tool now includes the PR URL in its output so the agent can share the direct link.
  • read_agent output now includes inbound messages that triggered each turn in multi-turn agents.
  • Autopilot task_complete summary is now required and renders as markdown.
  • Subagent elapsed time in /tasks view now freezes when idle and resumes when active.
Was this useful?

OpenAI Codex CLI

Sources Release notes → rust-v0.115.0 NOTES

Lightweight coding agent that runs in your terminal

Codex CLI gains full-resolution image inspection, realtime transcription, filesystem RPCs, Smart Approvals guardian routing, and a new Python app-server SDK.

└──▷ GET THIS VERSION
$ git clone --branch rust-v0.115.0 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.115.0
└──▷ USE IT
Override the OpenAI base URL (e.g., to point at a self-hosted or proxy endpoint) without changing the built-in provider.
toml
openai_base_url = "https://my-proxy.internal/v1"
  • Adds full-resolution image inspection: models can now opt into detail: "original" via view_image and codex.emitImage(..., detail: "original") for precision visual tasks.
  • Exposes codex.cwd and codex.homeDir inside js_repl, and persists codex.tool(...) / codex.emitImage(...) references across cells.
  • Adds a dedicated transcription mode for realtime WebSocket sessions, plus v2 handoff support via the codex tool, with a unified [realtime] session config block.
  • New v2 app-server filesystem RPCs cover file reads, writes, copies, directory operations, and path watching.
  • New Python SDK for integrating with the v2 app-server filesystem API.
+4 moreshow less
  • Smart Approvals can now route review requests through a guardian subagent in core, app-server, and TUI, reusing the same guardian session across follow-up approvals.
  • App integrations now use the Responses API tool-search flow, can suggest missing tools (tool_suggest), and fall back gracefully when the active model lacks search support.
  • Adds openai_base_url config override for the built-in provider.
  • The multi-agent wait tool is now consistently named wait_agent, aligning with spawn_agent and send_input.
└──▷ BREAKING ON UPGRADE
  • !The multi-agent wait tool is renamed to wait_agent; any configuration or prompts referencing the old name will break.
Was this useful?

Alibaba Qwen Code

Sources Release notes → v0.12.4 NOTES

Qwen Code v0.12.4 adds bundled /review and docs-audit skills, doubles the default output token limit to 16K.

└──▷ GET THIS VERSION
$ git clone --branch v0.12.4 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.12.4
└──▷ TRY IT
Run an instant code review on the current repository without any setup or custom skill configuration.
$ /review
  • Adds bundled /review slash command skill for out-of-the-box code review with no configuration required.
  • Adds docs audit and update helper skills for documentation quality workflows.
  • Increases DEFAULT_OUTPUT_TOKEN_LIMIT from 8K to 16K, enabling longer model responses by default.
  • Keeps rejected plan content visible in plan mode so users can reference prior proposals.
  • Auto-detects max_tokens from the model when not explicitly set by the provider.
Was this useful?
◆  AI Agent Frameworks

LlamaIndex

Sources Release notes → v0.14.18 NOTES

LlamaIndex v0.14.18 aligns text-match filters across vector backends and expands Bedrock Claude context windows to 1M tokens.

└──▷ GET THIS VERSION
$ git clone --branch v0.14.18 https://github.com/run-llama/llama_index.git
# already have the repo? check out this version:
$ git checkout v0.14.18
  • Aligns text match filters across llama-index-core and vector store backends for consistent filter behavior (#20883).
  • Sets context window size to 1M tokens for Claude Opus 4.6 and Sonnet 4.6 in llama-index-llms-bedrock-converse.
Was this useful?
Other / Uncategorized
◆  VECTOR DB RAG

LanceDB

Sources Release notes → v0.27.0 2 RELEASES · 2026-03-16 NOTES STABLE

LanceDB v0.27.0 adds a Rust expression builder API, fast_search parity, parallel inserts, and num_deleted_rows reporting.

└──▷ GET THIS VERSION
$ git clone --branch v0.27.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.27.0
└──▷ USE IT
Check how many rows were removed after a delete operation using the new num_deleted_rows field.
rust
let result = table.delete("status = 'inactive'").await?;
println!("Deleted {} rows", result.num_deleted_rows);
  • Adds expression builder API in Rust (feat(rust): add expression builder API) for type-safe query filters via Expr in query construction.
  • Supports Expr in projection queries in Rust, enabling type-safe column selection.
  • Accepts RecordBatch and Vec<RecordBatch> directly in create_table() and Table.add() in Rust, removing the need to wrap in a record-batch reader.
  • Adds num_deleted_rows field to the delete operation result, making it possible to inspect how many rows were removed.
  • Adds fast_search keyword argument parity between vector search and FTS search.
+10 moreshow less
  • Supports prewarm_index and prewarm_data on remote tables.
  • Adds support for remote index params via the remote SDK.
  • Allows passing Azure client/tenant ID through the remote SDK.
  • Supports dict-to-SQL struct conversion in Python table.update(), enabling structured updates without manual SQL string construction.
  • Supports field and data-type input in the Node.js add_columns() method.
  • Enables parallel inserts for local tables, improving write throughput.
  • Checks for dataset updates in the background, reducing latency for consistency-sensitive reads.
  • Shows reranker info in the hybrid search explain plan, making it easier to debug ranking pipelines.
  • Infers JS native arrays automatically in the Node.js binding.
  • Upgrades lance dependency to v3.0.0-rc.3, including bindings for fast_search.
└──▷ BREAKING ON UPGRADE
  • !create_table() and Table.add() in Rust now accept RecordBatch and Vec<RecordBatch> directly; callers previously relying on the old input types will need to update their call sites.
1 more release in this issue · 2026-03-16
python-v0.30.0 NOTES STABLE

LanceDB python-v0.30.0 adds fast_search parity, parallel inserts, expression-builder filters, and more new query and storage capabilities.

└──▷ GET THIS VERSION
$ git clone --branch python-v0.30.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.30.0
└──▷ USE IT
Run a hybrid search and confirm which reranker is active by inspecting the explain plan.
python
plan = (
    table.search('security breach', query_type='hybrid')
    .explain_plan(verbose=True)
)
print(plan)
  • Adds num_deleted_rows field to delete operation results, letting callers confirm how many rows were removed.
  • Adds fast_search keyword argument parity between vector and FTS searches so both query types share the same interface.
  • Adds expression builder API in Rust for type-safe query filters, replacing raw SQL strings in filter clauses.
  • Adds Expr support in projection queries (Rust), enabling type-safe column selection.
  • Adds add_columns() support for field/data type input in the Node.js SDK.
+11 moreshow less
  • Supports dict-to-SQL struct conversion in table.update() for Python, letting callers pass plain dicts for struct fields.
  • Adds bindings for fast_search via Lance 3.0.0-rc upgrade, enabling accelerated ANN lookups.
  • Supports prewarm_index and prewarm_data on remote tables to reduce cold-query latency.
  • Adds support for remote index params, extending index configuration to LanceDB Cloud tables.
  • Enables passing Azure client/tenant ID through the remote SDK for Azure-backed deployments.
  • Enables parallel inserts for local tables, improving bulk-write throughput.
  • Adds background dataset-update checks so stale reads are detected without blocking query threads.
  • Shows reranker info in hybrid search explain plans for easier pipeline debugging.
  • Caches schema of remote tables to reduce round-trips on repeated queries.
  • Infers JS native arrays automatically in the Node.js SDK, removing manual type hints.
  • Upgrades napi-rs from v2 to v3 in the Node.js binding layer.
└──▷ BREAKING ON UPGRADE
  • !create_table() and Table.add() in the Rust SDK now accept RecordBatch and Vec<RecordBatch> directly; callers passing other input types must update to these forms.
Was this useful?
◆  AI OBSERVABILITY

Langfuse

Sources Release notes → v3.159.0 NOTES

Langfuse v3.159.0 adds blob storage sync status badges, prompt label bulk selection, and ingestion event replay.

└──▷ GET THIS VERSION
$ git clone --branch v3.159.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.159.0
  • Adds 'Select all N / Clear' bulk-selection links and a combined search-and-create input for custom prompt labels in prompt management.
  • Shows a sync status badge and error alert in blob storage settings so operators can see integration health at a glance.
  • Adds ingestion event replay (v2) to re-process previously ingested events.
  • Adds a documentation link for backfilling on observation-level LLM-as-a-judge evaluations.
Was this useful?
my-toolchain — 0 tools
paste an install list to detect your tools

A brew list, a Brewfile, requirements.txt, a Dockerfile — or just the product names, free-form. Nothing leaves your browser.

    browse all tools →