Backfill issue -026 — The AI Toolchain

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   # 11 tools matched

AI & LLM Tooling

◆ AI Agent Frameworks

Agno (formerly Phidata)

v2.8.2 NOTES

Agno v2.8.2 adds FileSystem, a durable per-agent persistent filesystem with pluggable DB/local backends.

└──▷ GET THIS VERSION

$ git clone --branch v2.8.2 https://github.com/agno-agi/agno.git
# already have the repo? check out this version:
$ git checkout v2.8.2

›Adds FileSystem, a new durable agent filesystem primitive giving agents a private, persistent filesystem with pluggable DB or local backends and fail-closed per-user namespace isolation.

Was this useful?

PydanticAI

v2.17.0 NOTES

PydanticAI v2.17.0 adds arbitrary fields to usage types and caches OTel serialization to eliminate O(n²) instrumentation cost.

└──▷ GET THIS VERSION

$ git clone --branch v2.17.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v2.17.0

›Adds support for arbitrary fields on RequestUsage and RunUsage to accommodate extended provider-specific pricing and metadata.
›Caches per-message OpenTelemetry serialization, eliminating O(n²) instrumentation overhead for long conversation traces.

Was this useful?

◆ AI Coding Agents

Anthropic Claude Code

v2.1.219 NOTES

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Claude Code v2.1.219 adds Opus 5, sandbox network controls, depth-3 subagent nesting, and richer MCP diagnostics

└──▷ GET THIS VERSION

$ git clone --branch v2.1.219 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:
$ git checkout v2.1.219

└──▷ USE IT

Lock sandbox commands to only your approved hosts, preventing any unapproved outbound traffic without a prompt.

json

{
  "sandbox": {
    "network": {
      "strictAllowlist": true
    }
  }
}

Set the dynamic workflow size guideline centrally from a shared settings file instead of relying on each user's /config.

json

{
  "workflowSizeGuideline": "medium"
}

›Adds sandbox.network.strictAllowlist setting to deny all non-allowlisted hosts for sandboxed commands without prompting.
›Adds workflowSizeGuideline settings key so the advisory dynamic workflow size guideline can be set from any settings file; the /config row is hidden while one is active.
›Adds mcp_server_errors field to the headless stream-json init event, listing --mcp-config entries skipped by config validation; terminal runs also print a startup warning.
›Adds DirectoryAdded hook that fires after /add-dir or the SDK register_repo_root control request registers a new working directory mid-session.
›Subagents can now spawn nested subagents up to depth 3 (was 1); set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 to disable nesting.

+6 moreshow less

›Adds nested subagent forwarding in stream-json: subagents spawned at depth 2+ now appear when --forward-subagent-text is set, keyed by their spawning Agent tool_use id.
›Adds Claude Opus 5 (claude-opus-5) as the new default Opus model with 1M context window, priced at $10/$50 per Mtok in fast mode.
›Adds HTTP status and error text to claude mcp list and /mcp when a server fails to connect, plus a warning for MCP config values with hidden leading or trailing whitespace.
›Adds structured failure categories to self-hosted runner spawn and session failures, distinguishing hook errors, runner crashes, and config errors.
›Changes dynamic workflows to default to a medium size guideline (fewer than 15 agents); alternate sizes or unrestricted mode are selectable via Dynamic workflow size in /config.
›Removes Opus 4.7 from fast mode; /fast now applies to Opus 5 and Opus 4.8.

└──▷ BREAKING ON UPGRADE

!Opus 4.7 is removed from fast mode; /fast now applies only to Opus 5 and Opus 4.8.
!Managed MCP allowlist/denylist ${VAR} entries now resolve from the startup environment and managed-settings env instead of the settings-file env, which may change which values are substituted.

Was this useful?

Cline

v4.0.11 2 RELEASES · 2026-07-24 NOTES STABLE

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

Cline v4.0.11 adds Claude Opus 5 across six providers and Moonshot Kimi K3 support.

└──▷ GET THIS VERSION

$ git clone --branch v4.0.11 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout v4.0.11

›Adds Claude Opus 5 model support across Anthropic, Claude Code, Bedrock, Vertex, Cline, and OpenRouter providers, including 1M context window variants.
›Adds Moonshot Kimi K3 model support.

1 more release in this issue · 2026-07-24

desktop-v0.0.4 NOTES STABLE

Cline desktop v0.0.4 adds workspace-free chat, inline image attachments, one-time routines, and agentic compaction by default.

└──▷ GET THIS VERSION

$ git clone --branch desktop-v0.0.4 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout desktop-v0.0.4

›Headless routines now default to YOLO mode, enabling fully unattended execution without prompts.
›Agent sessions now use agentic compaction by default, keeping long conversations within model context limits more intelligently.
›Supports workspace-free chat sessions — start chatting without opening a project folder.
›New first-run onboarding flow on launch, with a setting to replay the new-user experience at any time.
›Drag-and-drop file attachment directly onto the chat transcript.

+4 moreshow less

›Image attachments now render inline in the chat transcript.
›Schedule one-time routines (in addition to recurring ones), with navigation to jump to a specific routine's run.
›New custom overlay title bar with in-app navigation.
›Redesigned channel setup as expandable cards.

Was this useful?

Charm Crush

v0.87.0 NOTES

Glamourous agentic coding for all

Crush v0.87.0 adds MCP OAuth 2.1, Claude Channels MCP extension support, and a system-wide config at /etc/crush/crush.json.

└──▷ GET THIS VERSION

$ git clone --branch v0.87.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.87.0

›Adds --channels opt-in flag to enable MCP channel servers (Claude Channels MCP extension), allowing MCPs to push notifications to Crush on events.
›Loads system-wide configuration from /etc/crush/crush.json, enabling NixOS and multi-user deployments to share a common Crush config.
›Supports OAuth 2.1 authorization for HTTP MCP servers (including SSE transport), unblocking MCPs that require OAuth.
›Detects the claude/channel MCP capability and routes incoming channel messages through Crush's event broker.
›LSP performance boost: filters LSP servers before searching $PATH, significantly reducing CPU and memory usage in large repositories.

+1 moreshow less

›Bang mode performance boost: eliminates quadratic shell output rendering, improving streaming throughput.

Was this useful?

OpenAI Codex CLI

rust-v0.146.0-alpha.8 3 RELEASES · 2026-07-24 NOTES CODE PRE-RELEASE

Lightweight coding agent that runs in your terminal

Codex CLI alpha adds ent26 enterprise plan support, MCP OAuth proxy routing, ephemeral thread forks, and workspace plugin publish capability.

└──▷ GET THIS VERSION

$ git clone --branch rust-v0.146.0-alpha.8 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.146.0-alpha.8

›Adds canPublishToWorkspace nullable metadata to plugin share contexts and plugin/share/save responses, letting clients surface workspace-directory publishing only when the capability is available.
›Supports excludeTurns: true in thread/fork to create ephemeral forks from paginated thread history without creating a rollout path or exposing the fork via thread/list.
›Routes MCP OAuth discovery and authentication status through each server's runtime HTTP client so MCP servers behind configured proxies are reliably discovered; caps local discovery at five seconds while preserving explicit login timeouts.
›Adds recognition of the ent26 enterprise plan across authentication, account protocol, backend rate-limit payloads, and app-server schemas, treating it as an enterprise workspace plan for cloud-config eligibility and usage-limit guidance.

2 more releases in this issue · 2026-07-24

rust-v0.146.0-alpha.7 NOTES CODE PRE-RELEASE

Codex CLI gains Agent Plugins manifest support, MCP runtime prewarming, and background MCP refresh coordination.

└──▷ GET THIS VERSION

$ git clone --branch rust-v0.146.0-alpha.7 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.146.0-alpha.7

›Supports Agent Plugins 1.0 schema: recognizes root plugin.json files and maps their portable metadata, skills/, and mcp.json into Codex plugin manifests, with .codex-plugin/plugin.json as a fallback overlay.
›Adds WaitForEnvironmentToolConfig as thread extension data, letting hosts override the model-visible wait_for_environment tool and environment_id descriptions.
›Adds background MCP runtime prewarming — a per-session worker eagerly rebuilds dirty MCP state after configuration, authentication, or explicit refresh changes, coalescing queued refreshes.
›Refreshes MCP runtimes automatically when session auth changes (auth mode, account details, or token), and rebuilds the initial MCP projection if authentication changes during session startup.
›Adds explicit RefreshMcpServers behavior that forces reconnection to every configured MCP server rather than reusing existing connections.

+4 moreshow less

›Refreshes MCP runtimes across thread startup and independently across active threads when managed MCP requirements change.
›Supports paginated thread forks.
›Exposes executor skills through skill tools and enables resource reads for explicit executor skills.
›Preserves output from hooks that exit before reading stdin (fast-exiting hooks no longer lose stdout when stdin is not consumed).

rust-v0.146.0-alpha.6 NOTES CODE PRE-RELEASE

Codex CLI alpha.6 adds remote code-mode host support via --code-mode-host, skill catalog context warnings, and nonblocking TUI turn interrupts.

└──▷ GET THIS VERSION

$ git clone --branch rust-v0.146.0-alpha.6 https://github.com/openai/codex.git
# already have the repo? check out this version:
$ git checkout rust-v0.146.0-alpha.6

└──▷ TRY IT

Connect the app server to a remote code-mode host over a secure WebSocket instead of the default local transport.

$ codex-app-server --code-mode-host wss://example.com/code-mode --listen stdio://

›Adds --code-mode-host flag to codex-app-server, accepting a wss:// WebSocket URL to connect the app server to a remote code-mode host (invalid schemes such as http:// or bare ws:// are rejected at parse time).
›Adds WebSocket transport to the code-mode host, enabling remote code-mode host connections through the app server.
›Exposes Browser Use requirements and remote skill icon URLs through the app server for downstream consumers.
›Attributes command executions to trusted plugin scripts, with attribution preserved across command approvals.
›Routes exec-server HTTP and WebSocket traffic through the configured proxy policy.

+4 moreshow less

›Preserves refreshed Apps tools across MCP runtime updates and keeps side conversations open when switching threads.
›Enforces writer ownership for thread archive and deletion operations.
›Supports incremental replay of updated thread items.
›Honors the configured SQLite home directory consistently across all state consumers, including startup backfill, thread listing, and paginated history.

Was this useful?

Earendil Works Pi

v0.82.0 NOTES

AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI

Pi v0.82.0 adds constrained tool sampling, OAuth login for OpenRouter and Kimi Code, and session-aware streaming bash integrations.

└──▷ GET THIS VERSION

$ git clone --branch v0.82.0 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.82.0

└──▷ TRY IT

Inspect session context inside a bash tool script — useful for logging or branching logic based on which model and provider Pi is using.

$ echo "Session: $PI_SESSION_ID | Provider: $PI_PROVIDER | Model: $PI_MODEL | Reasoning: $PI_REASONING_LEVEL"

Authorize Pi to use OpenRouter without touching API key config files — useful for shared or ephemeral environments.

$ /login

›Adds Tool.constrainedSampling with prefer/require modes for strict JSON Schema sampling and OpenAI Lark/regex grammar variants, gated by supportsStrictTools and supportsGrammarTools capability flags, across OpenAI, Anthropic, Amazon Bedrock, Google Gemini, and Mistral.
›Exposes PI_SESSION_ID, PI_SESSION_FILE, PI_PROVIDER, PI_MODEL, and PI_REASONING_LEVEL as environment variables to commands run by built-in and factory-created bash tools.
›Adds streaming bash_execution_update RPC events for direct RPC bash commands, correlated with request IDs.
›Adds OpenRouter OAuth PKCE login via /login, minting a user-controlled API key without manual configuration.
›Adds Kimi Code subscription OAuth login for the Kimi For Coding provider via /login, including device authorization and automatic token refresh.

Was this useful?

Alibaba Qwen Code

v0.21.0 NOTES

Qwen Code v0.21.0 adds QWEN_SYSTEM_IDENTITY_MD override, MCP force-reconnect, native video in /learn, @ session references, and configurable image-gen models.

└──▷ GET THIS VERSION

$ git clone --branch v0.21.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.21.0

└──▷ TRY IT

Override the agent's core identity for a custom deployment persona without modifying source files.

$ export QWEN_SYSTEM_IDENTITY_MD=/path/to/my-identity.md
qwen

Pass a video file to /learn so the agent can ingest and reason about recorded demos or walkthroughs.

$ /learn /recordings/demo.mp4

›Adds QWEN_SYSTEM_IDENTITY_MD environment variable to fully override the core system identity/prompt.
›Adds a usage statistics environment override via feat(cli): add usage statistics environment override (new env var for telemetry control).
›Supports forced MCP reconnects in serve, allowing clients to re-establish dropped MCP connections without restarting the daemon.
›Adds renderChatHeader slot to the web-shell component for injecting a custom session header.
›Supports native video input in the /learn CLI command.

+22 moreshow less

›Adds @ syntax for referencing prior sessions in the CLI, with completion tabs for discoverability.
›Adds configurable image generation models in core settings.
›Supports custom skill directories via CLI settings.
›Adds workspace selector button with add/switch dropdown in the composer toolbar.
›Shows subagent sessions in the web-shell detail panel.
›Adds rendered file previews in the web-shell.
›Adds workspace agent management UI in the web-shell.
›Adds a git mode selector when creating new sessions in the web-shell.
›Adds selective Shadow DOM isolation for web-shell component embedding.
›Adds workspace-level content generation in the serve layer.
›Persists workspace channel configuration across restarts.
›Adds daemon transport to the Java SDK (sdk-java).
›Emits the OAuth login URL as a single OSC 8 hyperlink in the terminal.
›Adds Goal v3 state protocol in core.
›Adds explicit channel delivery in the daemon.
›Scopes channel lifecycle to workspace runtimes.
›Auto-reruns a CI check that died on infrastructure failure, once (autofix).
›Auto-updates a PR when the merge gate rejects it due to a stale base (autofix).
›Posts PR review bodies bilingually when the PR description is in Chinese.
›Runs channel loops in daemon workers.
›Adds bounded Goal evidence verification in core.
›Propagates trusted daemon invocation context to child agents.

Was this useful?

◆ Local LLM Runtimes

llama.cpp

b10105 NOTES

llama.cpp b10105 consolidates --mmap, --mlock, and --direct-io into a single --load-mode argument

└──▷ GET THIS VERSION

$ git clone --branch b10105 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10105

└──▷ TRY IT

Lock model weights into RAM to prevent paging during inference on a memory-constrained host.

$ llama-cli --model model.gguf --load-mode mlock -p 'Explain quantum entanglement'

›Replaces the separate --mmap, --mlock, and --no-mmap flags with a unified --load-mode argument that accepts modifier-based values (e.g. mmap, mlock, dio) controlling how model weights are loaded from disk.

└──▷ BREAKING ON UPGRADE

!The previous --mmap, --no-mmap, and --mlock flags are now deprecated; combining them with the new --load-mode flag triggers a deprecation warning and may produce undefined behaviour — users should migrate to --load-mode.

Was this useful?

Other / Uncategorized

◆ AI OBSERVABILITY

Arize Phoenix

arize-phoenix-v19.6.0 2 RELEASES · 2026-07-24 NOTES STABLE

Phoenix 19.6.0 adds span_ids filter to get_spans in Python and TypeScript clients, plus OTLP JSON export for selected spans/traces in the UI.

└──▷ GET THIS VERSION

$ git clone --branch arize-phoenix-v19.6.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v19.6.0

└──▷ USE IT

Fetch only the spans you care about by passing specific span IDs — useful when you already know which spans to investigate and want to skip a broader query.

python

spans = client.get_spans(span_ids=["abc123", "def456"])

›Adds span_ids filter parameter to get_spans in the Python and TypeScript clients, enabling targeted retrieval of specific spans by ID.
›New UI capability to download selected spans or traces as OTLP JSON directly from the spans/traces view.

1 more release in this issue · 2026-07-24

arize-phoenix-v19.5.0 NOTES STABLE

Phoenix 19.5.0 adds evaluation metrics aggregation APIs, online agent trace evals with tool_count_per_turn and user_friction, a toxicity eval template, and Gemini 3.6 Flash support.

└──▷ GET THIS VERSION

$ git clone --branch arize-phoenix-v19.5.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v19.5.0

›Adds evaluation metrics aggregation APIs for programmatic rollup of eval results across traces.
›Adds online trace evals for agent workflows with two new built-in metrics: tool_count_per_turn and user_friction.
›Exposes dataset example source span via the API, linking dataset examples back to their originating trace span.
›Adds a toxicity gallery eval template designed to be input/output-agnostic for benchmarking.
›Adds Gemini 3.6 Flash and 3.5 Flash-lite model support in the Playground.

+1 moreshow less

›Adds semantic PXI tool icons to the UI for clearer agent tool visualization.

Was this useful?

◆ VECTOR DB RAG

LanceDB

v0.32.0-beta.3 4 RELEASES · 2026-07-24 NOTES STABLE

LanceDB v0.32.0-beta.3 adds block size configuration for full-text search indexing.

└──▷ GET THIS VERSION

$ git clone --branch v0.32.0-beta.3 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.32.0-beta.3

›Adds block size configuration for full-text search (fts) index building.

3 more releases in this issue · 2026-07-24

python-v0.35.0-beta.3 NOTES STABLE

LanceDB adds block size configuration for full-text search indexes.

└──▷ GET THIS VERSION

$ git clone --branch python-v0.35.0-beta.3 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.35.0-beta.3

›Adds block size configuration for full-text search (FTS) indexes, enabling tuning of index storage granularity.

python-v0.36.0-beta.0 NOTES STABLE

LanceDB python-v0.36.0-beta.0 adds remote branch diff/merge APIs, distributed plan metrics, and pre-Haswell wheel support.

└──▷ GET THIS VERSION

$ git clone --branch python-v0.36.0-beta.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout python-v0.36.0-beta.0

›Adds remote branch diff and merge client APIs for managing data branches programmatically.
›Publishes lancedb-compat wheels targeting pre-Haswell x86_64 hosts, enabling deployment on older CPU architectures that lack AVX2 support.
›Supports distributed analyze plan metrics in clients for observability into query execution across distributed setups.
›Extends skill references to work with jobs, including server connection contexts.

v0.33.0-beta.0 NOTES STABLE

LanceDB v0.33.0-beta.0 adds distributed query plan metrics, pre-Haswell wheel support, and remote branch diff/merge client APIs.

└──▷ GET THIS VERSION

$ git clone --branch v0.33.0-beta.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:
$ git checkout v0.33.0-beta.0

›Adds remote branch diff and merge client APIs for programmatic branch management.
›Publishes lancedb-compat wheels for pre-Haswell x86_64 hosts that lack AVX2 support.
›Supports distributed analyze plan metrics surfaced to clients for query performance observability.

Was this useful?

// END OF TRANSMISSION

The AI Toolchain · curated daily · July 24, 2026
Releases & features from the open-source and commercial AI tools that practitioners actually run.

The AI Toolchain — issue -026, July 24, 2026