The AI Toolchain — issue -034, July 16, 2026

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

Cline v4.0.9 adds GPT-5.6 ChatGPT subscription model support.

$ git clone --branch v4.0.9 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout v4.0.9

• ›Adds GPT-5.6 ChatGPT subscription models as selectable providers.

Glamourous agentic coding for all

Crush v0.85.0 adds a first-class question tool with single/multi-choice, free-text, and yes/no forms, plus an elapsed-time timer and LM Studio vision support.

$ git clone --branch v0.85.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.85.0

• ›Adds a question tool supporting single-choice, multiple-choice, free-form text, and yes/no question types, with the ability to chain questions into forms — includes full mouse support, mouse scrolling, paste support in text areas, and keyboard shortcuts for yes/no.
• ›Adds an elapsed-seconds timer so you can see how long a running LLM task has been executing.
• ›Wires LM Studio vision capabilities to SupportsImages, enabling image-input workflows with LM Studio models.
• ›Auto-retries authentication transparently when it fails, via OnAuthRefresh integration.

Copilot CLI v1.0.71 adds /voice device selection, canvas support, plugins marketplace subcommands, /worktree+/move split, and configurable subagent depth controls.

$ git clone --branch v1.0.71 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:
$ git checkout v1.0.71

{
  "githubMcpToolsets": ["repos", "issues"],
  "githubMcpTools": ["create_pull_request", "list_issues"]
}

$ copilot plugins marketplace browse
copilot plugins marketplace add <marketplace-url>

• ›Adds plugins marketplace subcommands to list, add, remove, browse, and update plugin marketplaces.
• ›Persists GitHub MCP toolset and tool configuration via settings.json keys githubMcpToolsets, githubMcpTools, etc.
• ›Adds /voice devices command to choose and persist the microphone used for voice mode.
• ›Adds subagents.maxDepth setting (up to 128) to control maximum sub-agent nesting depth; default lowered from 6 to 4.
• ›Adds a configurable maximum for Ctrl+R command history.

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

Pi v0.80.10 adds Kimi Coding adaptive thinking support with max level and empty-signature thinking block replay for K3.

$ git clone --branch v0.80.10 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.80.10

• ›Kimi Coding models now use adaptive thinking correctly; K3 exposes its supported max thinking level and supports replaying empty-signature thinking blocks.

$ git clone --branch v0.80.9 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.80.9

$ git clone --branch v0.80.8 https://github.com/earendil-works/pi.git
# already have the repo? check out this version:
$ git checkout v0.80.8

$ pi update --models

Qwen Code v0.19.11 adds /learn, archived session export, DingTalk DMs, general.notificationMode, and MCP workspace management.

$ git clone --branch v0.19.11 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.19.11

$ /learn

• ›Adds general.notificationMode config key to silence per-approval notifications selectively.
• ›Adds /learn command for user-initiated skill creation directly from the CLI.
• ›Changes the default approval mode from default to auto.
• ›Adds archived session export via the CLI (feat(cli): Add archived session export).
• ›Adds stateless generation SSE endpoint to the daemon for streaming without session state.

Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

CrewAI 1.15.3 adds step interception points, an @on hook dispatcher, and declarative flow support in the TUI.

$ git clone --branch 1.15.3 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:
$ git checkout 1.15.3

from crewai import on

@on('after_llm_call')
def inspect_llm_output(context):
    print(context)

• ›Adds step interception points and a generic @on hook dispatcher for wiring execution-boundary hooks at precise points in agent and crew execution.
• ›Adds an organization_id parameter to the PlusAPI client for multi-org API targeting.
• ›Runs declarative flows on the TUI with a headless terminal fallback.

PydanticAI v2.11.0 exports HistoryProcessor and adds actionable hints to usage-limit and tool-retry errors.

$ git clone --branch v2.11.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v2.11.0

from pydantic_ai import HistoryProcessor

• ›Exports HistoryProcessor from the public API, making it directly importable for custom conversation-history handling.
• ›Adds actionable hint messages to usage-limit and tool-retry errors, surfacing guidance at the point of failure.

Get up and running with Kimi-K2.6, GLM-5.2, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

Ollama v0.32.1 improves Gemma 4 tool calling, MLX timeout support, and agent context awareness.

$ git clone --branch v0.32.1 https://github.com/ollama/ollama.git
# already have the repo? check out this version:
$ git checkout v0.32.1

• ›MLX text model loading now respects the OLLAMA_LOAD_TIMEOUT environment variable, enabling timeout control for MLX backends.
• ›The interactive agent now receives the current working directory, giving models better project context during sessions.
• ›Agent web search and fetch now prompt users to run ollama signin when authentication is required.
• ›Improved Gemma 4 tool calling and multi-turn reasoning, including more reliable tool-response continuations.

An MCP that lets AI tools securely connect to your infrastructure, write IaaS code, debug issues, and assist during incidents - without risking production stability. Built for security teams to approve and infrastructure teams to experience like magic.

Emisar v0.29.0 tightens MCP argument coercion, sharpens error messages, and trims run summary payloads.

$ git clone --branch v0.29.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:
$ git checkout v0.29.0

• ›MCP now coerces scalar arguments sent as canonical strings — e.g. "limit": "50" or "dry_run": "true" — instead of failing the call, while true type mismatches name the field and the JSON type it was sent as (e.g. limit must be a JSON integer from 1 to 100; it was sent as a string.).
• ›MCP argument fault messages now name the offending field in every case: malformed refs advise copying the exact value from a prior tool result, enum faults list allowed values, and unknown arguments are spelled out against the allowed set.
• ›MCP run summaries drop zero-information fields: streams with no bytes omit stdout/stderr, byte-count, and truncation fields; output_complete appears only when false; and per-stream SHA-256 digests are removed from the MCP projection (they remain on the run detail page and audit record), shrinking a typical terse summary by about a third.
• ›Disabling two-factor authentication now requires a fresh step-up challenge, preventing a hijacked session from quietly weakening an account.
• ›Production session cookies now always carry the Secure flag.

$ git clone --branch v0.28.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:
$ git checkout v0.28.0

$ emisar state check-dispatch-log

$ QUARANTINE_DISPATCH_LOG=1 bash install.sh

$ emisar doctor

$ git clone --branch v0.27.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:
$ git checkout v0.27.0

$ git clone --branch v0.26.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:
$ git checkout v0.26.0

llama.cpp server gains text-only slot save/restore support for multimodal (mtmd) sessions

$ git clone --branch b10045 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10045

• ›Adds text-only slot save/restore capability to the server when running with multimodal (mtmd) support, enabling KV-cache slot persistence for text portions of multimodal sessions.

$ git clone --branch b10043 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10043

$ git clone --branch b10042 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10042

$ git clone --branch b10037 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10037

Composio experimental adds EveProvider with native defineTool support, step resolvers, and middleware hooks for Tool Router calls.

$ git clone --branch @composio/[email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:
$ git checkout @composio/[email protected]

• ›Adds EveProvider, which makes session.tools() return eve-native defineTools for direct integration with the Eve runtime.
• ›Adds defineComposioTools as a replay-safe step.started resolver for use with EveProvider.
• ›Supports (ctx, next) middleware hooks on Tool Router meta-tool calls, enabling callers to rewrite, deny, or transform requests before they execute.
• ›Preserves successful local-tool results when the remote half of a mixed COMPOSIO_MULTI_EXECUTE_TOOL batch fails at the transport layer, so callers can inspect which side effects completed before retrying.

$ git clone --branch @composio/[email protected] https://github.com/ComposioHQ/composio.git
# already have the repo? check out this version:
$ git checkout @composio/[email protected]

import { ComposioBlockedInternalUrlError } from '@composio/core';

try {
  await composio.files.upload('http://169.254.169.254/latest/meta-data/');
} catch (err) {
  if (err instanceof ComposioBlockedInternalUrlError) {
    console.error('Blocked SSRF attempt:', err.message);
  }
}

import { assertSafeFileUploadPath, isBlockedSensitiveFileUploadPath, BUILTIN_FILE_UPLOAD_PATH_DENY_SEGMENTS } from '@composio/core';

// Throws if path matches a blocked segment (e.g. ~/.ssh/id_rsa)
assertSafeFileUploadPath('/home/user/.ssh/id_rsa');

// Or check without throwing
if (isBlockedSensitiveFileUploadPath('/etc/passwd')) {
  console.warn('Blocked sensitive file upload');
}

Langfuse v3.218.0 adds eval job configurations to core data S3 export and forces EVENTS export source on events_only self-hosted deployments.

$ git clone --branch v3.218.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.218.0

• ›Includes eval job configurations in the core data S3 export, so evaluation setup is captured alongside trace data.

$ git clone --branch v3.216.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.216.0

$ git clone --branch v3.215.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v3.215.0

Weave v0.53.2 adds agent span feedback, eval metadata stamping, conversation signal filtering, server-side ingest sampling, and Base64/Data URL content ref conversion.

$ git clone --branch v0.53.2 https://github.com/wandb/weave.git
# already have the repo? check out this version:
$ git checkout v0.53.2

• ›Adds SubAgent.start_subagent() method and nests LLM/Tool/SubAgent spans under Turn and SubAgent in the Python SDK for structured agent trace hierarchies.
• ›Adds column_visibility field to ComparisonViewDefinition to control column display in comparison views.
• ›Adds scorer_trace_id field denormalized onto feedback records, linking scorer traces directly to feedback entries.
• ›Adds first-class SDK feedback for agent spans and turns, enabling programmatic feedback submission on agent traces.
• ›Adds agent eval span fields and stamps eval metadata onto agent spans for evaluation traceability.

Weaviate v1.38.5 adds a structured where-filter to the MCP hybrid search tool and warns when usage collection cycles overlap.

$ git clone --branch v1.38.5 https://github.com/weaviate/weaviate.git
# already have the repo? check out this version:
$ git checkout v1.38.5

• ›Exposes a structured where-filter on the MCP hybrid search tool, enabling filtered hybrid search through the MCP interface.
• ›Adds a warning when usage collection cycles overlap the configured collection interval, surfacing capacity issues in usage accounting.