v0.4.43

Changes

• kimi 25 (plus instant toggle alias) @evalstate (#646)

Validated Kimi-K2.5 with hugging face inference providers via novita. Structured, Vision, Tools and reasoning switch. Alias for "instant" added (moonshots term) -- use with kimi25?instant=on to disable thinking.

v0.4.42

fast-agent 0.4.42

• OpenAI verbosity setting through model string and TUI/ACP (e.g. Gpt-5.2?verbosity=high&reasoning=low)
• GLM 4.7 reasoning control (e.g. glm?reasoning=off)
• Agent Lifecycle Hooks (see skill from /skills add)
• Anthropic Structured Output support (supports reasoning with structured outputs. supply structured=tool_use to force legacy behaviour).
• Fix #644 Azure headers.
• Fix ruamel dependency for main package (from hf-inference-acp).
• Improve error handling/logging for tool hooks.

What's Changed

• Bump virtualenv from 20.35.4 to 20.36.1 by @dependabot[bot] in #638
• Feat/agent start hook by @evalstate in #639
• allow both tool_use and "proper" structured outputs for anthropic by @evalstate in #641
• refactor cpd (with opus post change) by @evalstate in #642
• Dev/0.4.41 by @evalstate in #643
• Dev/0.4.42 by @evalstate in #645

Full Changelog: v0.4.40...v0.4.42

fast-agent 0.4.40 (the self modifying version)

fast-agent 0.4.40

This release note wraps up a huge number of features and improvements made over the last couple of weeks. I hope you enjoy them.

fast-agent skills

Skills for modifying and working with fast-agent are available by default through /skills add. Use /skills registry to add a registry or select preconfigured (Hugging Face / Anthropic). Repo is here

Tool Hooks and Agent Lifecycle Hooks

Extend Agents with simple Python code using the inbuilt hooks. They let you hook Agent/Subagent lifecycle and LLM Loop events,
intercepting and modifying commands, adjusting inputs and outputs, starting containers, designing your own compaction strategies.

The following hooks are supported, extendable with simple Python code that gives you access to the execution context.

Tool hooks:

• before_llm_call
• after_llm_call
• before_tool_call
• after_tool_call,
• after_turn_complete

Agent Lifecycle hooks:

• on_start
• on_shutdown

When hooks are presented they are indicated with a ◆ next to the message or tool call.

Check the bundled skill to begin customising fast-agent! Switch on hot reloading with the --watch flag.

To create a subagent at runtime, simply use /agent <filename> --tool .

Example

Agent Card frontmatter:

name: dev-agent
model: codexplan
 tool_hooks:
   after_turn_complete: hooks.py:save_history

Python code

 from fast_agent.hooks.hook_context import HookContext
 from fast_agent.mcp.prompt_serialization import save_messages

 async def save_history(ctx: HookContext) -> None:
     save_messages(ctx.message_history, "history.json")

User Interface Improvements (a3 design)

fast-agent is unique in being CLI first - integrating perfectly with your terminal and preserving the scrollback buffer.  

fast-agent is has had a visual refresh, improving usability in long tool loops and workflows. Vertical space is much better used, and the stream is optimised for supervising LLM, Tool and subagent streams - highlighting important elements and actions, and enabling interventions where needed. Long gone are the days of monitoring single MCP tools calls!  Forms (human_input and MCP Elicitation) are also cleaner, with more content able to placed on the screen.

tool_stream.mp4

Of course, if you prefer a more refined front-end experience try Toad. Simply run toad acp "fast-agent-acp".

Integrated Shell

Run shell commands by entering ! and the command. Shell completion is supported (CTRL+SPACE). Enter ! to start a new shell, and exit to return to fast-agent.

shell2.mp4

Agent Query

Use # to send a message to an agent, and have the result returned to your input buffer ready for editing. Perfect for getting different model/agents feedback on plans, or with the human_input tool to create an interactive experience - or simply to get concise inputs.

hash_one.mp4

Bash output truncation

By default, shell output is truncated to 5 lines (configurable with output_display_lines under shell_settings in config).

To review the complete output use the /history review command.

Sessions and History Management

/history rewind lets you go back to a previous user turn, and place the message at that point in to the input buffer for editing.

Sessions are saved automatically and resumable with the /session resume <sessionid> command or --resume cli flag, and can be forked with /session fork. Use /session title <title> to give them memorable names.

fast-agent environments

Agent environments (config, skills and agent cards) can be specified with the --env <directory> flag. They can also be deployed as MCP Servers (fast-agent serve or used with any ACP client (fast-agent-acp shortcut).

Reasoning Adjustment

Model reasoning effort can be adjusted with the /model reasoning command, and supplied as a request parameter on the model string.

For example:

model: opus?reasoning=1024
model: codexplan?reasoning=xhigh
fast-agent --model haiku?reasoning=0

Existing configuration options and formats work, but more model settings will be migrated to this format over time.

OpenAI Codex Plan support

Use fast-agent auth codexplan to log in to your Codex CLI plan, and then specify the codexplan model to use it (currently aliased to GPT-5.2-codex). Use of the plan is denoted by a $ symbol in the status bar. The apply_patch tool has been ported to Python from Codex - improving token efficiency when using this model.

<img width="640"  alt="image" src="https://github.com/user-attachments/assets/99463eac-6591-43e6-8b76-ab3e928e8b6e" />

Aliyun Improved Model Support

Qwen3-Max added -- more soon

---

Big thanks to @iqdoctor !

What's Changed

• openresponses with sdk workarounds by @evalstate in #617
• Bump pyasn1 from 0.6.1 to 0.6.2 by @dependabot[bot] in #618
• Dev/0.4.32 by @evalstate in #619
• Fix/false llm error by @evalstate in #620
• Dev/0.4.35 by @evalstate in #623
• Dev/0.4.36 by @evalstate in #625
• Route stdio console output to stderr before stopping progress by @iqdoctor in #626
• Dev/0.4.37 by @evalstate in #628
• Dev/0.4.38 by @evalstate in #629
• Dev/0.4.39 by @evalstate in #630
• Bump aiohttp from 3.13.2 to 3.13.3 by @dependabot[bot] in #636
• Bump azure-core from 1.37.0 to 1.38.0 by @dependabot[bot] in #635
• Bump urllib3 from 2.6.2 to 2.6.3 by @dependabot[bot] in #634
• Bump filelock from 3.20.0 to 3.20.3 by @dependabot[bot] in #637
• Patch tool by @evalstate in #633
• style: format command_handler imports by @evalstate in #632

New Contributors

• @dependabot[bot] made their first contribution in #618

Full Changelog: v0.4.31...v0.4.40

v0.4.31

Open Responses and GPT-5.2-codex.

fast-agent has Open Responses Client support. This is in beta until official SDKs etc. are launched. This was added by fast-agent using gpt-5.2-codex which was launched in 0.4.30. The codex alias now points to this model.

https://fast-agent.ai/models/llm_providers/

Join the fast-agent Discord here to discuss.

• 5.2 codex by @evalstate in #616
  Full Changelog: v0.4.30...v0.4.31

v0.4.29

Responses and Hot Reload!

OpenAI Responses is now the default

Stateless Responses API (encrypted reasoning) is now the default for gpt-5 and o- series models. gpt-5.1-codex has been added as an alias.

Hot Reload of Agent Cards

By @iqdoctor -- hot reload of AgentCards (both Agents and Agents-as-Tools); completely dynamic, self-rewriting agents now possible. Available via TUI and ACP.

Other Changes

• ACP prompt sequencing by @evalstate in #615

Full Changelog: v0.4.27...v0.4.29

v0.4.27

What's Changed

Toad/Agent Card Examples

Added a new quickstart with hackable examples of Agents, Agents as Tools and Skills. Hot reload available (❤️ @iqdoctor ).

fast-agent: fast-agent quickstart toad-examples
From Toad: ctrl+o, Setup and /quickstart or use the wizard

Hash Commands

Send a message to an agent, and have it's result sent to the Input Buffer of the currently active agent.

MCP Client keep-alive

Client to Server ping healthcheck (thanks @Agent-Hellboy )

• server: add watch/reload for ACP and MCP by @iqdoctor in #594
• mcp tool progress notifications, mcp server hygiene by @evalstate in #601
• fix /skills slash command, fast-agent-acp system prompt by @evalstate in #604
• Fix exit traceback and watch tool files by @iqdoctor in #597
• Feat/hash command by @evalstate in #610
• Feat/toad enhance by @evalstate in #612
• Add optional client ping by @Agent-Hellboy in #611
• fix/minor changes, hf inference explorers demos by @evalstate in #613

New Contributors

• @Agent-Hellboy made their first contribution in #611

Full Changelog: v0.4.22...v0.4.27

v0.4.22

Agent Cards

New feature - Agent Cards.

Read the article by @iqdoctor here: https://github.com/evalstate/fast-agent/blob/main/plan/agentcard-standards-mini-article.md.

Agents can be loaded as peers, or as tools (using --card / --card-tool or /card <filename> [--tool] from ACP/UI. Cards are also auto-loaded from .fast-agent/agent-cards or .fast-agent/tool-cards, and can reference Python function tools.

• REPL. New AgentCard in md fromat and CLI loading/lazy hot swap with --agent-cards | --watch | --reload options. New workflow-md samples by @iqdoctor in #585
• Add command line option for loading card as tool by @evalstate in #590
• Feat/auto load by @evalstate in #593
• Add decorator function_tools support and RAG example by @iqdoctor in #592

Hugging Face Space Deployment

Improved support for HF Spaces deployment (deployment skill to be follow) - use HF OAuth to use end-user token for inference.

• Feat/hf outh by @evalstate in #595

What's Changed

• Fix/markdown stream improvements, uvloop by @evalstate in #582
• prompt toolkit hide user message by @evalstate in #583
• fix model resolution by @evalstate in #584
• Respect skills=None as explicit disable by @iqdoctor in #588
• change test running for speed by @evalstate in #587
• Feat/markdown experiment1 by @evalstate in #591

Full Changelog: v0.4.17...v0.4.22

v0.4.17

Sampling with Tools Release

• SEP-1577 - Sampling with Tools. @evalstate (#578)
• SEP-1036 - URL Elicitaitions.
• SEP-991 - CIMD support for OAuth.

Notes

You may need to remove earlier DCR tokens before re-authorizing. Use fast-agent auth to manage saved tokens.

v0.4.16

What's Changed

• Feat/skills manager by @evalstate in #567
• Feat/token display acp by @evalstate in #568
• Fix/types by @evalstate in #570
• fix diffs for non-streaming models by @evalstate in #571
• simplify streaming arguments by @evalstate in #572
• refactor to tool runner loop by @evalstate in #562
• minimax 2.1 and GLM 4.7 by @evalstate in #573

Full Changelog: v0.4.13...v0.4.16

v0.4.13

fast-agent 0.4.13 - What's Changed

MAKER Agent Type by @lucidprogrammer

MAKER (“Massively decomposed Agentic processes with K-voting Error Reduction”) wraps a worker agent and samples it repeatedly until a response achieves a k-vote margin over all alternatives (“first-to-ahead-by-k” voting). This is useful for long chains of simple steps where rare errors would otherwise compound.

https://fast-agent.ai/agents/workflows/#maker

https://arxiv.org/abs/2511.09030

Agents-as-Tools by @iqdoctor

https://fast-agent.ai/agents/workflows/#agents-as-tools

The Agents As Tools workflow takes a complex task, breaks it into subtasks, and calls other agents as tools based on the main agent instruction.
This pattern is inspired by the OpenAI Agents SDK Agents as tools feature.

Agent Skills Installer / Refresh to Standard

The /skills command now lets you add, remove and install skills from valid registries. Hugging Face and Anthropic are supplied by default, and can be overriden in the configuration file. Seen here with Toad. Some small tweaks to the existing skills implementation to bring it in line with https://agentskills.io/home.

Other Changes

Also thanks to @iqdoctor for collecting community resources here: https://fast-agent.ai/ref/fast_agent_ai_articles/. Let us know if there are more to add 🔥.

• fix gpt-4o document modalities
• add gemini 3 flash support
• other improvements for hf-inference-acp (model selection etc.)

Detailed Change List

• Review ACP implementation with new SDK Union types by @evalstate in #549
• Improve Google provider: Vertex AI (ADC) support & preview model fallback by @iqdoctor in #550
• Make agents ACP context aware by @evalstate in #551
• feat: add @fast.maker decorator for reliable multi-step workflows by @lucidprogrammer in #548
• Feat: hybrid Agents-as-Tools/MCP-as-tools experimental agent by @iqdoctor in #515
• fix(google): default maxTokens per Gemini model by @iqdoctor in #554
• Feat/hf acp by @evalstate in #555
• docs(examples): align agents_as_tools_simple with docs by @iqdoctor in #553
• Fix missing Hugging Face agent without token by @evalstate in #557
• update mcp library,gpt-4o pdf fix by @evalstate in #558
• Add model list display to set-model command by @evalstate in #556
• Add inference provider lookup for models by @evalstate in #559
• fix(hf-inference-acp): always set model and show provider info by @evalstate in #561
• Review MCP Server connection in agents ACP by @evalstate in #563
• Include full tool args in ACP titles by @iqdoctor in #565
• Implement agent skills standard support by @evalstate in #564
• gemini 3 flash by @evalstate in #566

Full Changelog: v0.4.4...v0.4.13