Help & Guides

Agents are the intelligent workhorses of the Alkimi platform, designed to transform how you interact with your data. Unlike static search tools, an Agent is a fully customizable AI entity that you configure to perform specific roles, ranging from a Socratic tutor guiding students through complex curriculum to a compliance officer verifying internal documents. Each agent is defined by its unique combination of knowledge, personality, and capabilities.

We call them "Agents" rather than just "Assistants" because they possess a higher degree of agency and capability. While an assistant might simply answer questions, an Alkimi Agent is engineered to actively process information, use tools, and reason through problems on your behalf. This choice also reflects our roadmap of constant improvement, as we continuously equip them with more automated capabilities and integrations. Grounded securely in your proprietary data, they don't just retrieve information; they understand it, apply it, and act upon it to drive your workflows forward.

Agent List

The Agents page is your central hub for viewing, managing, and creating AI agents within your workspace. It displays all agents you have access to in either a grid (cards) or list (table) layout.

• Agent Cards: Each card shows the agent's name, a snippet of its instructions, the model it uses, its credit cost per message, and when it was last updated.
• Search: Filter agents by name or description.
• Show all: By default, you only see the agents you have access to. Workspace owners (members with both the Create Agents and Manage Settings workspace permissions) can toggle it to also list agents in the current workspace they have not been granted access to; those appear greyed out and cannot be opened.
• Grid / List Toggle: Switch between visual cards and a compact table view with sortable columns (Agent, Model, Features, Cost, Updated).
• Right-Click Menu: Right-click any agent for quick actions: Settings, Save as Template, Export, Duplicate, or Delete.
• Hover Details: Hover the credit badge to see a cost breakdown (Message Generation + Tool costs). Hover the model for an Intelligence radar preview.
• + New Agent: The dropdown offers Blank Agent (opens the creation dialog), Import from JSON, From Collection (when the workspace has knowledge collections), and From Canvas Course (when the Canvas integration is enabled).

Creating an Agent

When you create a new agent, you have four ways to get started: using the AI Agent Builder, selecting a Template, starting from a Blank configuration, or Importing a JSON configuration file.

AI Agent Builder

The AI Agent Builder uses AI to configure your new agent for you. Simply describe what you want the agent to do in natural language.

The builder will analyze your request and automatically select the best template, model, and settings. It will even write the initial instructions for you. It grades the task with a tier rubric (Basic, Standard or Premium; it never assigns Free or Ultra) and recommends the Alkimi router for that tier, so the agent always uses that tier's best model. You can switch to a specific model afterwards on the Language Model page.

Templates

Alternatively, you can browse our catalog of pre-configured templates. These cover a wide range of use cases, from education to customer support.

After selecting a template, you'll be taken to the Finalize Agent wizard where you can review and customize the configuration before creating. The model list there shows the Alkimi routers for the template's recommended tiers, with the template's tier pre-selected; toggle Show all models if you want a specific model instead:

Agent Builder

The Agent Builder is a free, conversational helper for editing an existing agent (not to be confused with the AI Agent Builder, which creates new agents). Open it from the Agent Builder button in the page header while viewing any supported configuration tab (Profile, Chat, Language Model, Image Model, Instructions, Tools, or Knowledge), or from Quick Search when you are already on an agent page.

The assistant reads the agent's current settings, answers questions about models and capabilities, and returns edit proposals that you review before anything changes. Nothing is applied automatically: you accept each proposal explicitly, and applying one uses the same save path (and requires the same Manage Settings permission) as editing the configuration forms by hand.

How It Works

• Describe the change: Ask in plain language (e.g., "make the tone friendlier" or "enable web search"), or pick one of the starter prompts tailored to the tab you are viewing.
• Review proposal cards: Each proposed edit is shown as a card with a before/after diff. Related edits are grouped by the configuration page they affect.
• Accept or dismiss: Accept proposals individually, accept a whole batch, or dismiss the ones you don't want.
• Undo: Revert an applied proposal with one click, available while the live value still matches what was applied.
• Show me: Jump to (and highlight) the exact field a proposal will change on the underlying settings page (desktop only).
• Model recommendations: When you ask which model or tier an agent should use, the assistant grades the agent's work with the same tier rubric the AI Agent Builder uses for new agents (verdicts on real work call for Premium; pure knowledge-base lookup is fine on Basic; everything else lands on Standard) and recommends the matching Alkimi tier router. Ask for a specific model or name a provider and it proposes a concrete model inside that tier instead. It never recommends the Free or Ultra tiers unless you ask for them.

The assistant understands how settings interact and will reconcile dependent changes for you. For example, enabling file uploads on a model whose Message Spend Limit is too low produces a coordinated proposal that also raises the limit.

What It Can Edit

The Agent Builder can propose changes across the agent's profile, chat experience, model and spend limit, instructions (including Context Studio sections, safeguards, and variables), tools, and linked knowledge collections. Proposals go through the same validation as the manual configuration forms, so it can never apply a configuration you couldn't save yourself.

Availability & Limits

• Cost: Free to use (a 1-credit hold is placed on each message and released without charge), but you need at least 1 available credit to send a message.
• Permissions: Anyone who can view the agent's settings can chat with the assistant; applying proposals requires the Manage Settings permission (Owner or Editor role).
• Conversation memory: The assistant works from a rolling window of the most recent 16 messages, so long conversations keep working but older turns fall out of its context.
• Mobile: On smaller screens the panel opens as a full-screen overlay, and "Show me" is hidden because the settings page is not visible behind it.

Profile

This is where you define the agent's core identity and appearance. Everything users experience inside a conversation (greeting, message limits, chat features) lives on the separate Chat tab. The profile page is divided into several sections:

• Identity: Set your agent's Agent Name. The action buttons here let you quickly Duplicate the agent (linked knowledge collections are not copied), or use Import/Export to manage configurations as JSON files.
• Branding: Customize how your agent looks when embedded or shared. You can select a primary Theme Color, and the interface provides a real-time Preview of both the agent's and user's chat bubbles.
• Metadata: Access read-only system information, including the unique Agent ID, the agent's Dedicated URL, Created, and Last Updated timestamp.
• Danger Zone: A protected area to permanently Delete the agent and all of its associated history and data. Warning: This action is irreversible.

Chat

The Chat tab holds the Chat Experience card: everything that shapes what a user sees and can do inside a conversation with this agent.

• Greeting Message: The first message shown when a chat starts. Write it yourself or use the sparkle button to have AI generate one from the agent's instructions.
• Maximum Message Length: The most characters a user can type in one message, from 5,000 (the default) to 100,000. The default helps limit prompt-injection attempts; raise it when users need to paste long assignments or rubrics.
• Chat Features: Toggle Show Reasoning to reveal the agent's thought process (see the Enable Reasoning setting under Language Model for whether thoughts are generated at all), Enable Quoting so users can select part of a reply and ask about it, Allow Conversation Branching so users can edit an earlier message and explore an alternative thread, Enable Suggestions for follow-up questions, and Enable Math Keyboard for LaTeX input (with a per-category picker for which keyboard buttons appear).
• Reference Display Style: How knowledge and web citations are rendered in replies: Both Inline & Footer (numbered citations in the text plus a full list at the bottom), Inline Citations Only, or Footer List Only. Whether citations are produced at all is decided per source: per collection under Knowledge (References), and by Web Referencing under Tools.

Importing & Exporting Agents

The platform allows you to serialize your agent into a portable JSON format. This is perfect for:

• Backups: Saving snapshots of your agent's configuration.
• Sharing: Sending an agent template to a colleague.
• Versioning: Keeping track of different prompt iterations.

How to Use

Language Model

The Language Model page is where you select the model that powers your agent. The list opens with the Alkimi routers, which pick the best available model for you and keep the agent current as models change. Below them sit the individual models, grouped by provider. We recommend a router for almost every agent; pick a specific model only when you need that exact model.

• Model: Choose an Alkimi router (recommended) or a specific model (e.g., Gemini, Claude, GPT). Every entry is categorized by tier (Free, Basic, Standard, Premium, Ultra), and Alkimi Auto carries an Auto badge because it moves between tiers. Look for these icons in the list:
  • Image: Indicates multimodal support (the model can "see" images and documents).
  • Brain: Indicates reasoning support (the model can "think" through complex problems). It is either optional (transparent) or forced (opaque).
• Routing Preference: Shown only for routers. Choose Quality (the strongest option), Balanced (the default: best value among the strongest), or Cost (the most intelligence per credit). See Alkimi Routers.
• Message Spend Limit (Context Budget): Set the maximum "size" of each message in credits. This limit covers model-related costs including input context (history, knowledge, web content), reasoning, and output generation, but excludes separate tool operation costs.
  It cannot be set below the model's tier minimum (e.g., 4 for Basic, 12 for Standard) and has a fixed upper limit of 2,000 credits. When you switch models the limit is kept as-is and only raised if it falls below the new model's tier minimum. For Alkimi Auto, the limit also decides which tiers are in reach (see Routed Tiers below).
  JD

  MA301: Number Theory University of Alkimi

  Division Algorithm

  Agent 1-40+ cr/msg

  Profile

  Chat

  Language Model

  Language Model Settings

  Model Switching

  Model Details

  Image Model

  Instructions

  Capabilities

  Knowledge

  Tools

  Rules Coming Soon

  Management

  Permissions

  Statistics

  Distribution

  Canvas LMS

  Website Embed

  Division Algorithm - Language Model

  Test Chat Agent Builder

  Language Model Settings

  Select the underlying Large Language Model (LLM) that powers your agent. Different models have varying strengths, context windows, and capabilities (like image understanding). Learn more about credit costs.

  Language Model

  Alkimi Standard Standard 80

  Routing Preference

  Balanced

  Message Spend Limit

  Sets the maximum credits allowed per message. Higher limits enable longer conversations and more "memory", while lower limits keep costs predictable.

  Configured limit

  200 credits supports models up to Ultra

  Alkimi Standard's Max Limit

  292 credits

  model cap

  Standard 12Premium 40Ultra 200

  Tier minimum 12 credits ~182 A4 pages at this limit

  Actual cost starts at 1 credit per message. You are only charged for what you use (input + output), up to this cap.

  Enable Reasoning

  Controls whether the agent's thought process is generated.

  Enable File Uploads

  Allow users to upload images and PDFs in chats. Requires a compatible model and sufficient message spend limit.

  Max file capacity per chat: Up to 332 PDF pages/images (reserves up to 166 A4 pages from the spend limit).

  Cancel Save Changes

  Model Switching

  Let users pick a different model while chatting with this agent, and choose which models they can pick from. Switching is limited by the Message Spend Limit above.

  Allow Model Switching Mid-chat

  Let users select a different model while chatting with this agent.
• Routed Tiers: Shown only for Alkimi Auto and creator routers. Tick the tiers the router may use for this agent (for example, keep a homework helper off Premium). A tier above the spend limit is locked with a Raise limit shortcut, and the last reachable tier is Always on so the router always has somewhere to go.
• Settings: Enable features like Enable Reasoning (on by default; controls whether the agent's thought process is generated at all - whether users see it is the separate Show Reasoning toggle on the Chat tab. It can be turned off for any reasoning model, and models that cannot switch thinking off entirely instead run at the lowest reasoning effort they support with the thoughts hidden) and Enable File Uploads (allowing users to upload images and PDFs in chats - requires a model with image/PDF input support). The separate Model Switching card holds Allow Model Switching Mid-chat (letting users switch to a different model from the agent and model picker in the chat input area; the picker leads with the Alkimi tier routers and the agent's default, with every allowed model behind More models) and a Which models can users switch to? scope: every model in the same tier, the same model family, or an explicit list.

Alkimi Routers

An Alkimi router is a model entry that does not answer messages itself. Instead it always hands each message to the model that currently ranks best for the job, so your agent keeps improving as models are released, re-priced or retired without you touching its settings. Routers are billed exactly like the model they use, so a router never costs more than choosing that model yourself. They are the default choice for new agents and we recommend them unless you need one specific model.

• Alkimi Auto: The recommended default. For every message a fast routing step reads the request and picks a tier (Basic, Standard or Premium) that answers it well, so simple questions stay cheap and hard ones get a stronger model. It stays within the agent's Message Spend Limit and the tiers you allow, and it adds 1 credit for the routing step whenever more than one tier is in reach.
• Alkimi Basic, Standard, Premium (and Free, Ultra where enabled): Tier routers. Each one uses its tier's top-ranked model, ranked by the Alkimi Intelligence score and the agent's routing preference. If a cheaper paid tier's top model currently out-scores it with the same capabilities, the router quietly uses that better-value model instead and switches back as soon as its own tier leads again; paid tiers never fall through to Free. Choose one when you want a predictable cost per message but still want the best model in that price band.
• Creator routers (for example Google Auto): Pick a tier per message like Alkimi Auto, but only among one provider's models, preferring that provider's newest release in the chosen tier. Use them when your organization standardizes on a provider.
• Alkimi Image: The image-generation counterpart. Agents that generate images can select it as their image model to follow the platform's recommended image model automatically.

The Routing Preference dropdown tunes how a router trades quality against cost for this agent. Quality takes the strongest option in reach, Balanced (the default) takes the best value among the strongest, and Cost takes the most intelligence per credit. For routers that pick a tier per message (Alkimi Auto and the creator routers), it also sets the tier the router falls back to when it cannot decide: Quality answers on the highest tier in reach, Balanced on Standard (or the cheapest tier in reach if Standard is not), Cost on the cheapest tier in reach. The preference is stored per agent, so two agents on the same router can behave differently.

Routers respect your organization's model settings: if an administrator disables a model, every router that would have used it moves to the next best option for your organization. When no usable model is left for a router, it is hidden from the picker until one becomes available. For a router, the Model Details card shows the cost and capability figures of the model it currently uses (for Alkimi Auto, the most expensive tier it may use), without tying your agent to that model.

Because Alkimi Auto and the creator routers can land on different tiers, every model card and picker shows their Alkimi Intelligence as a range (a solid bar up to the weakest tier in reach, hatched up to the strongest) rather than a single score. Tier routers usually keep a single score; when their current model cannot read every attachment type and another model in the tier stands in for those messages, they show a range too. A tier router also accepts image and PDF attachments even when its current top model cannot read them: for that message it steps down to the best model in the tier that can, so agents on Alkimi Basic never reject a screenshot or a PDF because of the model of the day.

Model Details

The Model Details card provides a comprehensive technical breakdown of the selected model's performance, costs, and capabilities.

Alkimi Intelligence & Radar

The card leads with Alkimi Intelligence — a single percentile score (0–100) that summarizes where the model ranks among the models available on Alkimi. Models we do not offer are not part of the ranking. Below the score, an interactive radar chart visualizes performance across eight axes: Math, Science, Coding, Agentic, Reasoning, Long Context, Instruction Following, and Writing. Each axis shows the model's percentile rank; a P80 means the model beats 80% of the other available models on that axis. On every benchmark the best available model scores 100 and the weakest 0, so a model can only reach 100 overall by ranking first on every benchmark it is measured on.

Every axis combines at least one benchmark that covers small and open models, so cheaper models are not penalised for skipping frontier-only tests. The card states how many of the eight axes were actually measured ("7 of 8 categories"). An unmeasured axis is estimated from the five available models whose measured axes most resemble this model's (weighted by similarity), so the estimate reflects what comparable models actually scored there; an index built from fewer than four measured axes is labelled provisional.

When comparing models, values are color-coded relative to your current configuration: Green indicates a better value (e.g., higher benchmark or lower cost), Red indicates a worse value, and Yellow indicates a difference that is not directly comparable.

Capabilities

This section defines the operational limits and features of the model.

Credit Cost per Message

Understanding the credit consumption for your agent.

• Minimum Cost: The base cost to initiate a request (usually 1 credit).
• Tier Minimum: The minimum budget required by the model's tier (Basic: 4, Standard: 12, Premium: 40, Ultra: 200).
• Max Model Cost: The absolute maximum credits the model itself can consume for a single turn, based on your Message Spend Limit.
• Additional Tool Costs: Potential extra costs if tools (like Web Search) are used during the turn.

Benchmarks

We track key industry benchmarks across eight categories to help you compare model intelligence. Each category averages the percentile ranks of the benchmarks a model has been measured on. Not all models have scores for every benchmark — missing data is displayed as a dash.

Benchmark scores are sourced from Artificial Analysis, Context Arena, LiveBench, EQ-Bench and the Vectara Hallucination Leaderboard. Individual benchmark references: AIME, GPQA, SciCode, LiveCodeBench, Terminal-Bench, Humanity's Last Exam, MRCR v2, IFBench.

Message Spend Limit (Context Budget)

The message spend limit (technically known as the context budget) is a per-agent setting that controls the maximum credits available for each message. It determines both how much context your agent can access (conversation history, knowledge base content, and web results) and how it can generate its response (reasoning and final answer). Think of it as the "size" of each conversation turn.

How It Works

The spend limit directly affects two key aspects of your agent's performance:

• Input Context Size: Higher limits allow your agent to consider more conversation history, more content from your knowledge base, and more web search results when formulating its answer. This means better context awareness but higher potential cost per message.
• Output Generation: The limit also covers the cost of generating the agent's response, including its reasoning process and the final answer. More complex reasoning or longer responses will consume more credits.
• Intelligent Allocation: Our system automatically distributes your limit across different content sources (history, knowledge, URLs) based on relevance, ensuring the agent gets the most important information within your constraints.

Tier-Based Limits

Each model tier has a minimum required spend limit. This limit, set by the agent owner, acts as a maximum cost per message. The actual cost is variable based on usage (but never less than 1 credit). You must set a limit that meets the tier's minimum, and you can set it higher for more complex tasks.

• Free Models: No tier minimum; the limit still applies as the message's maximum cost.
• Basic Models: Require a minimum spend limit of 4 credits per message.
• Standard Models: Require a minimum spend limit of 12 credits per message.
• Premium Models: Require a minimum spend limit of 40 credits per message.
• Ultra Models: Require a minimum spend limit of 200 credits per message.

Note: The agent configuration interface will show you the available limit options for your selected model.

Alkimi routers follow the same rule. A tier router such as Alkimi Standard needs its tier's minimum (12 credits). Alkimi Auto needs only the Basic minimum (4 credits) and can use any allowed tier whose minimum fits under the limit, so raising the limit is how you let it reach Standard or Premium; the Routed Tiers control on the Language Model page shows which tiers are in reach.

Choosing the Right Limit

Consider these factors when setting your agent's spend limit:

• Task Complexity: Simple queries (like "What is X?") can work with lower limits, while complex analysis or multi-step reasoning benefits from higher limits.
• Knowledge Base Size: If your agent needs to search through large knowledge collections, allocate more limit to ensure relevant blocks aren't truncated.
• Conversation Length: For ongoing conversations where context from previous messages is important, higher limits help maintain continuity.
• Response Detail: If you need comprehensive, detailed answers with step-by-step reasoning, ensure the limit allows for longer outputs.
• Cost Management: Remember that every message consumes credits up to your set limit. For high-volume use cases, consider starting with a moderate limit and adjusting based on results.

You can adjust the spend limit at any time from your agent's Language Model page. We recommend experimenting with different limits to find the sweet spot for your specific use case. For detailed information on how these limits affect billing and credit consumption, see our Billing & Pricing documentation.

Image Model

The Image Model page controls whether and how the agent generates images inline in chat. Each generated image is billed separately from the message, on top of the message's own cost, and you only pay for images that are actually generated.

• Image Generation: Turn image generation on or off for the agent. When it is on, the Image Model picker chooses the model used; Alkimi Image follows the platform's recommended image model automatically.
• Allow Image Model Switching Mid-chat: Lets users pick a different image model from the agent and model picker. A scope setting limits the choice to all image models, selected pricing tiers, selected model families, or a hand-picked list.
• Image Quality and Image Resolution: Shown for models that expose those knobs. Auto lets the agent decide per image; otherwise you can pin a quality (Low, Medium or High) or a resolution (0.5K to 4K, depending on the model). Higher settings cost more credits per image, and the page shows the estimated range for the selected model.
• Max Images per Message: Caps how many images a single reply can generate (1 to 6, default 2), bounding the per-message image spend.

Instructions (Prompt Engineering)

The instructions are the set of instructions your agent follows. Effective prompt engineering is key to building a great agent. While you can write the entire prompt manually in the "Raw" editor, we recommend using the Context Studio, which provides a structured and intuitive way to build and manage your agent's instructions.

Note: Every new agent starts with a preloaded, general-purpose prompt, so it works out of the box. Templates are chosen when you create an agent (see Templates above); for an existing agent, use the Context Studio's Add from Library sections or the Agent Builder to customize the prompt. Instruction content (excluding safeguards) is limited to 10,000 characters.

The Context Studio

The Studio is designed to give you fine-grained control over your agent's behavior while providing a real-time preview of how your instructions will be formatted.

• Persona: This is the agent's core identity. A good persona is concise and sets the stage for all other instructions (e.g., "You are a helpful and witty customer support agent for a software company.").
• Tone of Voice: Select from a list of predefined tones to quickly shape your agent's personality and response style.
• Sections: These are the building blocks of your prompt. You can add, remove, and reorder sections to structure the agent's instructions. There are two types:
  • Freeform: A simple text area where you can write paragraphs of instructions.
  • Structured List: A hierarchical list editor that helps you organize complex rules or protocols into a clear, nested format.
• Add from Library: To speed up prompt creation, you can add pre-written, reusable sections from the library, covering common use cases like constraining the agent to its knowledge base or defining an escalation protocol.
• Variables: Use system variables (like {{agent_name}}) or define your own custom variables (like {{course_name}}) to make your prompts dynamic and reusable.
• Live Preview: As you build your prompt in the Studio, a live preview on the right shows you exactly what the final system prompt will look like, including how variables are resolved.

Safeguards

Safeguards are pre-built safety instructions that are automatically appended to your agent's prompt. They help ensure your agent behaves responsibly by enforcing boundaries around identity, data handling, content, and scope. Your organization's administrators can set default safeguards that are applied to all new agents (see Organization Policies), and you can further customize them per agent.

• Identity: Prevents the agent from impersonating real people or claiming false credentials.
• Data: Protects sensitive information like PII, credentials, or internal data from being leaked or misused.
• Content: Limits the generation of harmful, inappropriate, or unsafe content.
• Scope: Keeps the agent focused on its intended purpose and prevents it from straying off-topic.

You can manage safeguards from the "Manage Safeguards" button in the instructions editor. We recommend keeping safeguards to 5 or fewer to avoid consuming too much of the agent's context window.

Best Practices for Prompting:

• Be Specific: Clearly define the agent's role, the task it should perform, and any constraints. Instead of "Summarize text," try "Summarize the following text into three bullet points, focusing on the key financial outcomes."
• Provide Examples: Use the structured prompt editor to give few-shot examples of desired input and output. This is one of the most effective ways to guide the model's behavior.
• Define the Persona: Tell the agent how to behave. For example: "You are a helpful assistant. Your tone should be friendly and professional."

Knowledge

Connect your agent to one or more knowledge collections. This grounds the agent in your specific data, enabling it to answer questions based on the documents you've provided. An agent can access multiple collections simultaneously. See Knowledge Context & Cost below for more details on how connected collections affect performance.

• Recall Control (experimental): Adjusts how widely the agent searches before relevance filtering. Presets run from Exact (narrowest, lowest cost) through Strict, Balanced (recommended) and Lenient to Max Recall (widest, highest cost per message). Because relevance filtering decides what the model actually sees, the presets often change little for focused questions; the Context Estimate below shows their effect on your collections.
• Collection @Mentions: Users can always reference a specific collection in chat with @collection-name, which searches that collection for the current message on top of automatic retrieval. The setting decides how far mentions reach: only collections connected to the agent, or any collection in the workspace the user can access.
• Per-collection switches: Each connected collection has its own settings. @Mention Only keeps the collection out of background context unless it is mentioned. References lets the agent cite passages from it (adds 1 credit per message). Figures (on by default) lets the agent show figures and images extracted from its documents inline in chat; when off, the agent can only describe them in words. Downloads (off by default) lets the agent offer its files for download in chat; a user is only offered a download when they also have view access to the collection in the content explorer, and the check is repeated every time a download panel loads.

Knowledge Context & Cost

Everything the agent can search lives in one Connected Collections card with a single Save Changes button: the connected collection rows, an always-present Other workspace collections row that controls workspace-wide @mentions, and Recall Control at the bottom. Directly below it, the Context Estimate strip shows what connected knowledge does to a single message: the risk level, the credits knowledge adds per message, the share of the message's room it takes, and how many searchable blocks the always-active collections hold. Click Details for the full breakdown; it opens on its own whenever risk is raised.

The agent never pastes whole collections into a message; each answer draws on the blocks that pass relevance selection, and how many that is depends on the corpus and the question. Rather than assuming a fixed block count, the estimate predicts that range by running the real retrieval pipeline against the connected collections on three families of probe questions drawn from the corpus: focused lookups (the low end), everyday questions (the middle), and broad "summarize this document" requests (the high end). The prediction is computed in the background and stored, so opening the page normally shows it at once; the first visit after the corpus or retrieval settings change shows a brief Predicting... state that refreshes on its own. Only when prediction is not possible does the strip fall back to a simpler block-size model. The estimate runs over saved connections, so it appears after the first Save Changes.

The details show:

• Share of the message's room and added cost per message, each from a focused question to a broad one. Questions that ask for every match ("list all rows where...") can use up to the spend limit.
• Room for knowledge: whether that room is set by the agent's Message Spend Limit or by the model itself. A low spend limit leaves little room even on a large-context model.
• Blocks per answer: how many blocks a typical answer draws on, out of the searchable blocks across the always-active collections.
• Risk: Low when heavy questions still leave room for conversation history; Elevated when they take a large share of the room or many blocks are very large; High when heavy questions would push retrieved evidence out of the message or exceed what the model reads reliably. Each raised risk explains its cause and the setting that addresses it.
• Mentions: when some collections are connected as @Mention Only, a footnote shows what a turn that mentions them takes and costs, sized from those collections' blocks.

Connecting a collection, changing Recall Control, or ingesting new content re-runs the prediction in the background; model and spend-limit changes reprice it immediately. The strip is an estimate only; observed usage lives in the Where Credits Go card on the agent's Statistics page.

Tools

Extend your agent's capabilities beyond its core knowledge by enabling tools. Using these tools will consume credits from your account; for a detailed breakdown, please see our Billing & Pricing documentation.

Note: Web Search and URL Browsing are disabled by default and must be explicitly enabled for each agent; Web Referencing is on by default.

• URL Browsing: Enables the agent to "read" the content of a specific webpage when you provide a link in your message. Each URL browsed consumes credits.
• Web Search: Allows the agent to perform real-time web searches. Configure when it runs:
  • Disabled: The agent cannot perform web searches.
  • Explicit: The agent searches only when the current message contains a valid @web badge.
  • Implicit: The agent decides when a search is useful. A valid @web badge in the current message always forces a search, and a badge with sources also restricts the search to those sources.
• Web Referencing: Adds citations and footnotes for web search results and browsed URLs, so users can see which page a claim came from. Citations for knowledge blocks are switched on per collection (the References toggle under Knowledge); clicking one opens the source document at the cited page and highlights the block. How both kinds of citation are laid out is the Reference Display Style on the Chat tab.

LMS Tools

A second card on the Tools tab, LMS Tools, covers structured coursework that moves between chats and learning management systems:

• Import to Chat: Lets users pull enabled Canvas content (course overviews, modules, pages, files, quizzes and rubrics) into a conversation. Requires Enable File Uploads on the Language Model page and a connected Canvas account with a verified Teacher or Designer role; see Canvas LMS.
• Quiz / QTI: The agent can produce revisable quiz cards that download as official QTI or platform-specific files for Canvas, Moodle, Blackboard, D2L and other LMS platforms.
• Rubrics: The agent can produce revisable rubric cards, download or upload official Canvas rubric CSV files, and import cards directly into Canvas when connected.

Web Search Source Policy

When Web Search is enabled, source policy controls the starting scope and whether users may add other sources.

• Open: Users can choose any valid public domain, organization source, or built-in group. Default domains are selected for new plain web badges and also govern searches the agent starts in Implicit mode.
• Restricted: The allowed domains form a hard boundary. Users cannot add custom sources outside that scope, and search results and pages outside it are unavailable to the agent.
• Default domains: The enforced starting scope for a new plain web badge and an implicit search. In Restricted mode, every default must be inside the allowed scope.
• Allowed domains: Required in Restricted mode. If no defaults are configured, the allowed list becomes the effective scope for new badges and implicit searches.

A domain includes its subdomains. Where accepted, .edu and .gov are suffix filters. Choose 1–3 domains for a focused scope. Default lists, allowed lists, and user badges each have a maximum of 5 domains. Organization sources and built-in groups are suggestions, not restrictions, unless the agent's Restricted policy includes them.

Search and page-browse operations are billed when they execute. Use the current cost display on the Tools page and the Billing & Pricing documentation for current rates.

Note: File Uploads are configured on the Language Model page, not the Tools page. See the Uploading Files section for details on supported formats and capacity.

For detailed tips on how to get the most out of these tools, see the Using Tools section of the Chat documentation.

Rules

The Rules Engine is a powerful feature (currently under development) that allows you to define conditional rules to dynamically modify your agent's behavior. Based on triggers like user questions or specific conditions, you will be able to:

• Inject additional instructions on-the-fly.
• Include specific knowledge sources, overriding default search strategies.
• Force the use of a particular tool with custom instructions.

Permissions

Control who can interact with and manage your agent. Permissions are managed at both the organization level and the individual agent level. New agents start with No Default Access: only the creator (Owner) and the members you add explicitly can use them.

For a deeper dive into our access control model, please see our Security Documentation.

• Default Access: The All Workspace Members row sets the role for every member of the agent's workspace who has no explicit role. It can be Editor, Viewer, User, or No Default Access (Owner cannot be a default).
• Explicit Permissions: You can grant specific members a role for this agent that overrides the default.

Agent Roles

Agent roles are fixed and global — they cannot be customized or extended. Each member is assigned exactly one role per agent. This is different from organization and workspace roles, which are fully customizable.

Statistics

Gain insights into how your agent is performing, how users are interacting with it, and where your credits are being spent. The statistics page provides a comprehensive dashboard with real-time metrics.

Overview Metrics

The top row provides a quick snapshot of your agent's health and usage:

• Users: The total number of unique users who have chatted with your agent. Includes 7-day and 30-day active user counts.
• Chats: The total number of conversation threads. Tracks active vs. deleted chats.
• Messages: The volume of interactions. Useful for tracking engagement trends over time.
• Response Quality: Monitors how often your agent hits the Message Spend Limit. A high percentage of "Context Limit Hits" suggests you may need to increase your limit for better performance.
• Performance: Tracks the average latency (speed) of responses, broken down by "Time to First Token" (how fast it starts typing) and total generation time.

Charts & Trends

Visualize your agent's activity over the last 7, 30 or 90 days, or a custom date range:

• Activity & Engagement: Correlates message volume with unique users and chat sessions.
• Performance: Tracks response times to help you identify slowdowns or latency spikes.
• Credit Consumption: Shows daily credit usage to help you manage costs.
• Response Quality: Visualizes why generation stopped (e.g., natural completion vs. hitting limits).

Peak Hours Heatmap

Understand when your users are most active. This heatmap aggregates usage by day of the week and hour of the day (in your local time), allowing you to identify usage patterns and peak traffic windows.

Where Credits Go

This card shows what the messages in the selected period actually spent credits on, so you can see where the money goes and what to change. It is the observed counterpart of the Context Estimate on the Knowledge page.

• Composition bar: A stacked bar splits total credits across everything the model read (System scaffolding, your Instructions, conversation History, retrieved Knowledge, fetched Web pages, user Attachments), what it wrote back (Response), and Tools (web search, URL browsing, references, image generation). The empty tail of the bar is Open room: the share of the prompt window an average message left unused, so a short tail means the spend limit or model window is the constraint. Each segment lists its share and its average credits per message.
• Knowledge Context: How full the prompt window ran (and how often messages hit the spend limit), what share of the prompt retrieved knowledge took, what knowledge cost per message, and how often evidence had to be trimmed to fit. Figures are medians over the period.
• Before / After: When you change knowledge settings or connected collections inside the period, a comparison appears so you can see what the change did to knowledge cost.

Cost & Tool Breakdown

A detailed audit of tool spend:

• Tool Usage: Counts how many times the agent performed web searches or browsed specific URLs.
• Cost Breakdown: Splits total credit consumption between the core Model (thinking & generating) and external Tools (searching & browsing).
• Avg Cost per Message: A key metric for forecasting future costs as your user base grows.

When the Response Quality chart shows replies stopped at the maximum response length, the warning badge on each such message in chat says which limit cut it off: the agent's Message Spend Limit (raise the slider on the Language Model page) or the model's own output limit (a higher spend limit would not help; switch to a model with a larger output limit or ask for the answer in parts).

Activity

The Activity page provides a historical log of all changes made to a specific agent. This is useful for tracking how an agent's configuration has evolved over time and identifying which team members made specific updates.

• Configuration Changes: Updates to the agent's name, greeting, or instructional prompt.
• Model & Budget Updates: Changes to the underlying language model or the message spend limit.
• Knowledge Connections: When knowledge collections are added to or removed from the agent.
• Tool Toggles: Enabling or disabling capabilities like Web Search or File Uploads.
• Permission Updates: Changes to the explicit access roles granted to organization members.

Distribution

The distribution pages allow you to embed your agent as a chat widget on any website, share it directly via a dedicated link, or integrate it with a Learning Management System. This is perfect for customer support, website navigation, or educational tools.

Dedicated Page

The simplest way to share your agent is through its Dedicated Page. This is a standalone, full-page chat interface hosted by Alkimi.

• Direct URL: A unique link (e.g., /a/[agent-id]) that you can share via email, social media, or internal docs.
• Access Control: If Public Access is enabled, anyone with the link can chat (as a guest). If disabled, users must be logged into your organization.

Embed Methods

The Website Embed page offers two embed methods (the Canvas variant below lives on the Canvas LMS page):

1. Script Tag (Floating Widget)

Adds a floating chat bubble to the corner of your website. This is the most feature-rich option.

• Welcome Message: Customize the initial greeting shown to visitors.
• Prefetching: Enable "Aggressive Prefetching" to load assets immediately on page load for faster startup.
• Display Mode: Configure the widget to open as a floating window or a side panel, or allow users to toggle between them.
• Hidden FAB: Optionally hide the Floating Action Button (FAB) and open the widget programmatically.
• JavaScript API: Control the widget from your code using methods like openWidget(), closeWidget(), and toggleDisplayMode().

2. Iframe (Inline Embed)

Embeds the chat interface directly into a specific area of your page layout.

• Styling: Customize borders, rounded corners, and box shadows to match your site's design.
• Positioning: Choose between Inline (standard div), Floating (fixed position), or Fullscreen.

3. Canvas LMS Embed

A specialized fullscreen iframe designed for Learning Management Systems (like Canvas or Blackboard).

• Fullscreen: Automatically takes up 100% of the viewport.
• Borderless: Removes default borders and styling to blend seamlessly with the LMS interface as an external tool.

For deeper Canvas integration (automatic course embedding, per-module control), see the Canvas LMS Integration section below.

Interactive Tools (Coming Soon)

These tools create a dynamic, two-way interaction between your agent and the user on your website. When a user sends a message from your site, the agent can be given context about the page they are on and what they are looking at. In turn, when the agent responds, it can highlight elements on the page to guide the user's attention.

Note: These features are currently under development. Once available, they will be disabled by default for security and privacy, requiring explicit enablement in the agent's integration settings.

• Current Page Awareness: Allows the agent to see the webpage the user is currently on, providing valuable context for its responses.
• User View Mirroring: Lets the agent see what the user is interacting with, such as selected text. This helps the agent understand the user's focus without them having to explain.
• Interactive Highlighting: Grants the agent the ability to highlight elements on your webpage and scroll the user to them, making it an excellent tool for guiding users through your site.

Canvas LMS

If your organization has the Canvas integration configured, the Canvas LMS tab provides a native connection to your Canvas courses. This goes beyond the basic iframe embed by offering direct control over where your agent appears within Canvas.

• Course-Level Embedding: Browse your Canvas courses and embed the agent as a page within any course. Students can access the agent directly from their course navigation.
• Per-Module Embedding: For more targeted placement, embed the agent into specific course modules. This lets you place an AI tutor right next to the relevant learning materials.
• Embed Management: View which courses and modules currently have the agent embedded, and remove it from any location with a single click.

Note: Canvas integration requires your organization's administrator to configure the Canvas OAuth connection and enable the canvas feature flag. See your organization's Integrations page or contact your admin to set this up.

Discord

This feature is currently under development. Soon, you'll be able to connect your agent to Discord, allowing it to interact with users in your servers.

Slack

This feature is currently under development. Soon, you'll be able to connect your agent to Slack, allowing it to interact with users in your workspace.

Teams

This feature is currently under development. Soon, you'll be able to connect your agent to Microsoft Teams, allowing it to interact with users in your organization.