Everything ClawAI does
Nine model families, five routing modes, nine ways to put several models on one problem, memory that persists between conversations, files you can ask questions about, twelve workspace connectors, and a usage view that shows exactly where your allowance went.
Last reviewed 2026-07-25
Every frontier model family
One subscription reaches all of these. Switch between them inside a single conversation — there is nothing to install and no provider account to create.
Anthropic Claude
Careful reasoning over long documents, strong code review, and the most reliable instruction following in the roster.
Models
- Claude Opus 5
- Claude Sonnet 5
- Claude Fable 5
- Claude Haiku 4.5
OpenAI GPT
Broad general capability with excellent tool use and dependable structured output.
Models
- GPT-5
- GPT-5 mini
- GPT-4o
- o4-mini
Google Gemini
Very large context windows, fast responses, and native handling of images, audio and video.
Models
- Gemini 3 Pro
- Gemini 3 Flash
- Gemini 2.5 Flash
Moonshot Kimi
Long-context analysis and agentic work at a fraction of frontier pricing.
Models
- Kimi K2
- Kimi K2 Thinking
Zhipu GLM
Strong bilingual reasoning with an explicit thinking mode you can turn on for harder problems.
Models
- GLM-5.1
- GLM-5
- GLM-4.7 Thinking
Alibaba Qwen
Excellent code generation across a wide range of model sizes, from fast to very large.
Models
- Qwen3 235B
- Qwen3 80B
- Qwen3-Coder-Next
DeepSeek
Mathematics, algorithms and step-by-step derivations, with a dedicated reasoning variant.
Models
- DeepSeek V3.2
- DeepSeek R1 0528
xAI Grok
Fast conversational answers with awareness of recent events.
Models
- Grok 4
- Grok 4 Fast
Amazon Bedrock
AWS-hosted models for organisations that already run their workloads on Amazon.
Models
- Nova Pro
- Titan
- Bedrock-hosted Claude
The roster tracks each provider’s release schedule. When a new frontier model ships it appears in your model list — no plan change, no new contract, no second bill.
Five ways to choose a model
Auto is the default and handles most work well. The other four exist for when you want the decision yourself, or want it biased in a specific direction.
- Auto
- ClawAI classifies each message — code, reasoning, analysis, creative, image, everyday — and sends it to a model that is strong at that class.
- Manual
- Pin one model to the conversation. Every message goes to it until you change it.
- Low latency
- Prefer the fastest model that can answer, for quick back-and-forth work where waiting costs more than a slightly weaker answer.
- High reasoning
- Prefer the strongest reasoning models regardless of speed or cost, for problems where being right is the only thing that matters.
- Cost saver
- Prefer the cheapest model that can still handle the request, so a fixed allowance covers more work.
You always see the decision
Every answer carries the provider and model that produced it, the confidence score behind the routing choice, the reasons that drove it, and what it cost against your allowance. If a provider fails mid-request, the model that stepped in is recorded too.
Nine ways to use more than one model
A single answer from a single model is often enough. When it is not, these modes put several models on the same problem.
- Compare
- Send one prompt to up to five models at once and read every answer side by side, with per-model latency and token counts.
- Consensus
- Run the same prompt across several models and synthesise one answer from where they agree — and flag where they do not.
- Escalation
- Start on a fast, cheap model and move up to a stronger one automatically, but only when the answer falls below a quality threshold.
- Best-of-N
- Generate several candidate answers to the same prompt and keep the one that scores highest.
- Repair
- Fix specific defects in an answer you already have — a broken code block, a wrong figure — instead of regenerating the whole thing.
- Verify
- Have a second model check an answer for correctness before you see it, with a configurable limit on revision rounds.
- Role packs
- Run a prompt through a small team of role-specialised models — planner, critic, writer — each handing off to the next.
- Pipelines
- Chain several of these stages into one named workflow you can re-run on new input.
- Judge and Critic
- An independent model scores a response against explicit criteria; a Critic pass adds written feedback on what is weak and why.
The free plan includes one trial run each of Compare, Judge and Research. Paid plans carry monthly quotas that rise with the tier, and Team and above are uncapped.
Memory that carries between conversations
ClawAI remembers what is worth remembering — and shows you every item, so you can correct it.
- Memory records
- Facts, preferences, instructions and summaries, each stored with a category, a confidence score and a record of where it came from.
- Suggestions you approve
- Nothing is remembered silently. Candidates land in a queue you approve or reject, and only high-confidence non-sensitive items are auto-approved — at a threshold you set.
- Context packs
- Bundle reusable text, files, links and memory references into a named pack, then attach it to any conversation. Packs are versioned, so you can see what changed and roll back.
- Scopes
- Scope a memory to yourself, to one conversation, to a project or to a workspace, so work context does not leak into personal chats.
- Context receipts
- Every answer records which memories and pack items went into the prompt, in what order, and how much of the token budget each consumed.
- Controls
- Pause all memory, pause one item, set an expiry date, mark something sensitive so it is redacted, or delete it outright. Every change is written to an audit log.
Memory and context are switches on each conversation. Turn them off for a one-off question and the prompt contains nothing but what you typed.
Ask questions about your own files
Upload a document and ask about it. ClawAI handles extraction, chunking and retrieval so the model answers from your content rather than its training data.
- Upload
- PDF, DOCX, spreadsheets, CSV, JSON, Markdown, plain text, code files and images.
- Chunking and retrieval
- Files are split into passages and indexed, so only the parts relevant to your question are pulled into the prompt.
- Attach to any message
- Pick files per message, including in Compare runs, so several models can be asked about the same document at once.
- Per-model delivery
- Each model receives the form it handles best — a native image, a native PDF, or extracted text — and every message shows which form each model actually got.
- Scanned documents
- A scanned PDF with no text layer is run through OCR before it reaches a model, and flagged when the recognition confidence is low.
- Upload checks
- Every upload is virus-scanned, checked against its declared file type, screened for dangerous filenames, and rejected if an archive turns out to be a decompression bomb.
Uploads count against your plan’s file size and storage limits. Files are removed on a retention schedule, and you can delete any of them yourself at any time.
Connected to the tools you already use
Connect an account once over OAuth and ClawAI can search it, pull context from it into a conversation, and act on it when you approve the action.
GitHub
Repositories, issues, pull requests and code search.
GitLab
Projects, issues, merge requests and pipeline status.
Bitbucket
Repositories, pull requests and commit history.
Jira
Issues, sprints, comments and status transitions.
ClickUp
Tasks, lists and status changes.
Slack
Channels, threads and message history.
Gmail
Threads, search, and drafted replies you approve before anything is sent.
Google Drive
Documents and folders in Google Drive.
OneDrive
Files and folders in OneDrive.
SharePoint
Sites, document libraries and lists.
Confluence
Spaces, pages and attachments.
Figma
Files, frames and comments.
Connections sync on a schedule and on webhooks, so search results stay current. Credentials are encrypted at rest, scoped to your account and revocable in one click. How many connections you can hold depends on your plan.
Images and documents, not just text
ClawAI produces finished artefacts from a conversation, not only chat replies.
- Image generation
- Describe an image and get it back in the thread, with the prompt and settings kept alongside it.
- Document export
- Turn any answer, or a whole conversation, into a formatted file you can hand to someone else.
- Research runs
- Search the web, fetch and read the pages, and get an answer with the sources it came from rather than an unattributed summary.
Export formats
- DOCX
- CSV
- HTML
- Markdown
- TXT
- JSON
You can see what you are spending
Usage is not a black box. Every request is metered, attributed and visible.
- Usage dashboard
- Allowance consumed today and this month, broken down by model, with the remaining balance in the same units your plan is quoted in.
- Per-answer detail
- The model that answered, why it was chosen, how long it took, how many tokens it used and what it drew from your allowance.
- Audit log
- Logins, plan changes, connector activity, memory edits and generated content, each recorded with a timestamp and an actor.
- Live progress
- While a model works you see the stage it is in, the text as it arrives, its reasoning when the model exposes it, and running token and timing counters.
When you reach a limit you are told which limit, how much is left on the other windows and when it resets. Nothing is cut off silently.
How your data is handled
Mechanisms that exist in the product today, described plainly.
- Accounts and sessions
- Passwords hashed with Argon2, short-lived access tokens, and refresh tokens that rotate on every use so a stolen token is detectable.
- Role-based access
- Every account carries a role and an explicit permission set, checked in the interface and again on every backend endpoint.
- Credential encryption
- Provider and connector credentials are encrypted at rest with AES-256-GCM and are never returned to the browser.
- Rate limiting
- Per-account request limits protect both your allowance and the platform from runaway loops.
- Encryption in transit
- TLS from the browser to the edge, and TLS again between every internal service, with certificates verified at each hop.
- Service isolation
- Each service owns its own database and cannot read another’s. A failure in image generation cannot reach your conversations.
Requirements the hosted app cannot meet
ClawAI holds no compliance certifications today, and the hosted app sends requests to third-party model providers under their terms. If your organisation cannot accept that, a private deployment inside your own network — running open-weight models only — is the answer. We scope those individually.
Talk to us about a private deploymentTry it on the free plan
No card required. Send a few messages, run one Compare, and see whether the routing earns its keep.