Intent Router
Reads every turn and routes it — CRM lookup, knowledge answer, or action — with dynamic graded scoring. No keyword lists, nothing hardcoded.
Dynamic · graded
AgenTorQ AgenTorQThe same agentic core — Workmates, grounding, private memory, dynamic cards and 14+ models. Run it as an instant standalone workspace, or native inside Salesforce. Same brains — you pick the place to run it.
The managed package that installs from AppExchange and runs inside your org — native LWC + Apex, grounded on live CRM, governed by your own sharing model.
For teams already on Salesforce who want AI Workmates native to their Salesforce org.
The web app at agentorq.com — no install. Bring your own keys for 14+ AI providers, connect your apps, and start running Workmates in any browser.
For anyone who wants a powerful personal AI workspace — instantly.
A zero-trust, model-agnostic platform built into Salesforce — grounded on your records, documents and knowledge, taking real actions on 14+ AI providers with your own keys. Plus DevSpace Studio: an AI-native IDE for Apex, LWC, Flows and GitHub, custom MCP connectors, and Workmates that remember.
Every turn runs through a live, graded Intent Router — then the agent grounds on your CRM, retrieves from your knowledge, or takes a real action. Salesforce-native, zero-trust, fully traceable.
Reads every turn and routes it — CRM lookup, knowledge answer, or action — with dynamic graded scoring. No keyword lists, nothing hardcoded.
Dynamic · gradedLive reads from Salesforce records through a regex-free SOQL guard and DML guard — plus Gmail, Calendar, Drive & REST APIs. Real records, real time.
SOQL · connectorsRAG grounding over your folders, files and chunks with embeddings — PDFs, XLSX, OCR. Cited passages, never a hallucinated record.
RAG · embeddingsThe agent doesn't just talk. ACTs create & update records, run Flows, send email and render edit forms — every write checked against your access.
Write-back · safeSetup to runtime — every layer is configurable, governed and observable.
Most "AI for CRM" is a chat box wired to one vendor model with limited grounding. AgenTorQ is a real platform — open, grounded, zero-trust, and built to act.
OpenAI, Azure, Gemini, DeepSeek, Mistral, Cerebras, Perplexity, SambaNova, MiniMax, Groq, Grok (xAI), HuggingFace, AI21, Ollama — plus a generic OpenAI-compatible adapter. Switch per agent or per task. You bring the keys, so there's no per-conversation tax.
Reads live records, files and your knowledge base — then cites the sources.
Creates & updates records, runs Flows, sends email, shows forms.
Named PII is masked before anything leaves; every query passes a SOQL guard and writes a DML guard; secrets live in an encrypted external credential vault with per-user OAuth. Your data stays in your org — only the masked prompt reaches the model you chose.
XLSX, PDF, CSV & scanned images via extraction + OCR.
Open connectors with MCP auto tool-discovery and a connector registry.
Logs, audit trail, call logs and a live health console — no black box.
ACT cards, charts, markdown & Mermaid in a themeable console.
Beyond chat. AgenTorQ ships a working surface of agentic capabilities — configurable Workmates, private memory, real web search and interactive cards that actually do the work.
Configurable AI employees — give each a purpose, base instructions and guardrails, then attach its own knowledge and apps. AI-fill any field, or AI-fill the whole config in one click.
WorkmateToggle multiple Workmates active at once. They collaborate and stay grounded together on the same task — a team of specialists, not a single bot.
CollaborateEach Workmate remembers each user personally — and privately. Memory is isolated per user, so what it learns about you never leaks to anyone else.
PrivateReal, latest, dated headlines — on any engine, even non-browsing models. A server-side search keeps answers precise and current, with sources you can see.
CurrentThe AI proposes actions as interactive cards — multi-select bulk (“email these 16 leads” → one card, checkboxes, approve-all), record create/update forms, and live charts.
InteractiveType while it's still answering. Your next message queues and auto-sends the moment the answer lands — a fluid, Claude-style flow with no waiting.
FluidPDF, Word, Excel and CSV become clean text for every model; images use native vision where available, or OCR where it isn't. Drop a file and just ask.
Omnivore14+ AI providers with your own keys; Gmail, Outlook, Salesforce and 600+ apps via per-user OAuth. Your spend, your data, your connections.
OpenOrganizations with SFDC-style Profiles and per-member permissions, strict per-user isolation, owner protection and deny-by-default access — secure by design.
EnterpriseDevSpace Studio is a VS Code-style workspace inside AgenTorQ — three panes: explorer, tabbed editor, and a docked Dev Workmate that knows exactly what file and selection are open. Browse, read, edit and deploy Apex, LWC and Flows — then let it do the change end-to-end: describe it or select code → it edits → you see a diff → approve → deploy · test · fix.
Dev WorkmateIDE‑BRAINJump to any Apex class, LWC or Flow instantly with Ctrl/⌘ P — no scrolling the tree.
⌘ PSearch within the open file with Ctrl/⌘ F, match highlighting and jump-to-match.
⌘ FRun the sf CLI over your org or git over GitHub, right inside the studio.
sf · gitA test-coverage overlay in the editor gutter — see covered vs. uncovered lines at a glance.
OverlayLeap from a class to its test class (and back) in one move — no hunting.
1‑hopVisualize any Flow as a clean diagram — read the automation without opening Setup.
VisualizerAsk the Dev Workmate for a senior-level review — issues, risks and concrete fixes.
ReviewDescribe a change or select code — it edits, shows a diff, and on approve runs deploy · test · fix.
AgenticDevSpace is a real workspace tab — your panes, tabs and context survive a refresh.
StatefulEnforced operating modes, a multi-platform code editor, custom MCP connectors, 218 expert templates, per-Workmate memory and enterprise WorkSpaces — each with its own showcase below.
Five enforced modes, from careful to hands-off — switchable per turn, and honored strictly on the server.
A code-provider registry means the same studio edits your Salesforce org and your GitHub repos — with real git, no local clone.
A Perplexity-style “Add custom connector” flow — name it, give it an icon and an MCP URL, sign in once, and AgenTorQ discovers its tools automatically.
A gallery of 218 ready-made expert Workmates across 16 divisions — engineering, design, marketing, sales, product, security, finance, support, GIS, game-dev and more.
Each Workmate keeps its own memory — Concepts, Entities, Workstreams and Notes — and applies it on future turns, on top of cognee-backed semantic recall across sessions.
Run AgenTorQ alone, or as a WorkSpace: an admin invites users, builds SFDC-style Profiles from a feature-toggle grid, and assigns them — with per-user overrides.
Every capability, broken down to its finest detail — and exactly what each edition gives you. Both share the same engine; this is the honest, line-by-line map of where each one shines.
→ Swipe the table sideways to see both editions.
| Capability | AgenTorQ for SalesforceNATIVE MANAGED PACKAGE | AgenTorQ StandaloneWEB APP · AGENTORQ.COM |
|---|---|---|
| Workmate AI agents configurable AI employees | ||
| Create a Workmate | Workforce / Agent Builder, no code | Agents builder, instant |
| Set Workmate purpose | Purpose field | Purpose field |
| Base instructions & grounding | Section 0 base instructions | Editable per Workmate |
| Guardrails & constraints | Per-Workmate guardrails | Per-Workmate guardrails |
| Per-Workmate knowledge base | Per-agent RAG objects | Folders, files, chunks |
| Per-Workmate connected apps | Scoped connectors | Per-agent apps |
| Model per Workmate | Per-agent / per-task engine | Engine + model dropdown |
| AI-fill a single field | ✨ AI fill per field | ✨ AI fill per field |
| AI-fill all fields at once | AI fill all | AI fill all |
| Type-ahead model selector | Top-5 suggested | Searchable selector |
| Templates gallery | Dev / Admin / Business | 63 cross-industry |
| One-click create from template | Pre-filled, editable | Pre-filled, editable |
| Per-Workmate isolation | Isolated config & data | Isolated per agent |
| Autonomous run / scheduler | Heartbeat + queueable ticks | On-demand |
| Org read/write toggle | Native R/W, FLS-aware | Via connectors |
| Workmate teams multiple active, grounded together | ||
| Multiple Workmates active at once | Multi-agent swarm | Toggle multiple active |
| Team grounding on one task | Coordinated on a task | Grounded together |
| Agent-to-agent coordination | Swarm orchestration | Collaborate live |
| Toggle a Workmate on / off | Per-agent enable | One-tap toggle |
| Per-user private memory remembers you, isolated to you | ||
| Remember on every turn | Porting to package | Remembers each turn |
| Recall on send | Porting to package | Recall-on-send |
| Per-user memory isolation | Per-user store | Per-user dataset |
| Private per Workmate, per user | Per-user scope | Isolated & private |
| Knowledge-graph memory | — | Strong graph memory |
| Remember on KB add | RAG ingest | Auto-remembered |
| Remember on Workmate add | Config-scoped | Auto-remembered |
| Clear / reset memory | Admin-scoped | User-clearable |
| Document memory | RAG over files | RAG over files |
| Model-agnostic web search real, dated, on any engine | ||
| Live web search | Web-search control | Server-side search |
| Real, latest headlines | Augmented results | Real news headlines |
| Dated headlines | Source-dependent | Dated & current |
| Works on non-browsing models | Model-dependent | Any engine |
| Source citation on results | Cited sources | Source icons + links |
| News vs general modes | General augment | News & general |
| Precise & current answers | Grounded on CRM first | Precise & current |
| Universal dynamic cards actions, forms & charts | ||
| Single action, approve-run | ACT cards | Action card |
| Bulk items in one card | 360 bulk card | e.g. 16 leads, one card |
| Multi-select checkboxes | Row checkboxes | Row checkboxes |
| Select-all | Select-all | Select-all |
| Per-row run | Run a single row | Run a single row |
| Approve-all action | Approve-all | Approve-all |
| Per-row pass/fail status | Per-row ✓/✗ | Per-row ✓/✗ |
| Record create form | Create card | Create card |
| Record update form | Update card | Update card |
| Typed form fields | Typed inputs | Typed inputs |
| Picklist fields | Picklists | Picklists |
| Live chart | Charts in console | Live charts |
| Chart type switcher | Switch chart type | Common types |
| PPTX export | Roadmap | Export to PPTX |
| PNG export | Roadmap | Export to PNG |
| Mermaid diagrams | Mermaid render | Markdown-first |
| Markdown rendering | Rich markdown | Rich markdown |
| Message queuing Claude-style flow | ||
| Type while it's answering | — | Queue your message |
| Auto-send queued on completion | — | Auto-sends on land |
| Any file, any model attachments by type | ||
| PDF → text | Doc intelligence | Parsed to text |
| Word (DOCX) → text | Common formats | Parsed to text |
| Excel (XLSX) → text | XLSX intelligence | Parsed to text |
| CSV → text | CSV intelligence | Parsed to text |
| Plain text files | Supported | Supported |
| Image — native vision | Model-dependent | Native vision |
| Image — OCR for non-vision | OCR built in | OCR fallback |
| History sanitize of attachments | Sanitized | Sanitized |
| Works for every model | Text for any model | Text for any model |
| Models & BYOK 14+ providers, your keys | ||
| 14+ AI providers | All supported | All supported |
| OpenAI | Supported | Supported |
| Azure OpenAI | Supported | Supported |
| Google Gemini | Supported | Supported |
| DeepSeek | Supported | Supported |
| Mistral | Supported | Supported |
| Cerebras | Supported | Supported |
| Perplexity | Supported | Supported |
| SambaNova | Supported | Supported |
| MiniMax | Supported | Supported |
| Groq | Supported | Supported |
| Grok (xAI) | Supported | Supported |
| HuggingFace | Supported | Supported |
| AI21 | Supported | Supported |
| Ollama (self-host) | Self-hosted | Self-hosted |
| Generic OpenAI-compatible adapter | Any endpoint | Any endpoint |
| Bring your own keys | Your keys | Your keys |
| Switch model per task | Per task | Per task |
| Live model fetch (hundreds) | Live fetch | Live fetch |
| Free engine fallback to start | Porting in | Free engine to start |
| Streaming responses | Token streaming | Token streaming |
| Connectors your apps, per-user OAuth | ||
| Gmail | Per-user OAuth | Per-user OAuth |
| Outlook | Via connectors | Per-user OAuth |
| Google Calendar | Via connectors | Connected |
| Google Drive | Via connectors | Connected |
| Google Docs | Via connectors | Connected |
| Google Sheets | Via connectors | Connected |
| Salesforce (as a connector) | Native + connector | Connect your org |
| GitHub | Via connectors | Connected |
| 600+ apps via Nango | REST / MCP layer | 600+ catalog |
| Per-user OAuth | Per-user | Per-user |
| Reconnect a connector | Reconnect | Reconnect |
| Disconnect a connector | Disconnect | Disconnect |
| REST connectors | REST | REST |
| MCP connectors, auto tool-discovery | MCP discovery | Connector layer |
| Enterprise & security Organizations, Profiles, zero-trust | ||
| Organizations (workspaces) | Org RBAC | Organizations |
| SFDC-style Profiles | Profiles | Profiles |
| Per-member permission overrides | Overrides | Overrides |
| Join codes | Admin invite | Join codes |
| Promote member to admin | Promote | Promote |
| Members — manage / use | Granular perm | Granular perm |
| Profiles — manage | Granular perm | Granular perm |
| Connectors — manage / use | Granular perm | Granular perm |
| Workmates — manage / use | Granular perm | Granular perm |
| Knowledge base — manage / use | Granular perm | Granular perm |
| Engine keys — manage / use | Granular perm | Granular perm |
| Settings — manage | Granular perm | Granular perm |
| Analytics — manage / use | Org logs | Granular perm |
| Org-shared agents / KB / keys overlay | Org records | Auto-shared on add |
| Per-user isolation | Strict isolation | Strict isolation |
| Owner protection | Owner-protected | Owner-protected |
| Deny-by-default access | Deny-by-default | JWT deny-by-default |
| No privilege escalation | Enforced | Enforced |
| FLS / sharing (USER_MODE) | USER_MODE enforced | Connector scopes |
| Named PII masking | Masked before send | Scrutiny passes |
| SOQL / DML guards | Regex-free guards | N/A (no SOQL) |
| External credential vault | Encrypted vault | Encrypted keys |
| Audit trail & logs | Full audit | Activity logs |
| Zero-trust posture | Zero-trust | Zero-trust |
| Deployment & access where it lives | ||
| Where it runs | Inside your Salesforce org (LWC + Apex) | Any browser at agentorq.com |
| Install method | AppExchange managed package | None — just sign in |
| Live CRM grounding | Native, live records | Via Salesforce connector |
| Data residency | Stays in your org | Your keys; per-user memory |
| Sign-in gate | Salesforce auth | AgenTorQ sign-in required |
| Time to value | Master Setup, minutes | Instant, no setup |
✓ included · − partial / via connectors / on the roadmap · × not in this edition. Both editions run the same AgenTorQ engine; capabilities evolve release to release. AgenTorQ is an independent product and is not affiliated with or endorsed by Salesforce.
Every question is routed to the right source, masked for PII, run on the model you chose, and returned with citations.
Ask in plain language, get an answer grounded in live data — with the sources it used, a rich action card, and a builder to shape every capability.
AgenTorQDEEPSEEK · GROUNDEDThe native AI agent that ships with your CRM is convenient — but it's locked to one vendor's model, metered per conversation, and limited in what it can read and do. The honest side-by-side.
| Capability | AgenTorQOPEN AGENTIC PLATFORM | The built-in CRM agentNATIVE / VENDOR-LOCKED |
|---|---|---|
| Choice of AI model | 14+ providers + self-host + generic adapter, per task | Locked to the platform's own model family |
| Pricing model | Bring your own keys — no per-conversation tax | Metered per conversation / consumption credits |
| Document & OCR | XLSX, PDF, CSV & scanned-image OCR built in | Limited file understanding; OCR rarely included |
| Your knowledge base (RAG) | Folders → chunks, embeddings, per-agent RAG, cited | Constrained grounding within the vendor framework |
| Connectors | REST + MCP with auto tool-discovery — extend to anything | Closed ecosystem; approved integrations only |
| Zero-trust security | Named PII masking + SOQL/DML guards + credential vault | Data routed to the vendor model; controls vary |
| Takes real actions | Create/update, run Flows, send email, show forms | Limited to predefined topics/actions |
| Observability | Logs, audit trail, call logs & health console | Limited visibility into calls and failures |
| Branding & UI | Themeable console, ACT cards, charts, Mermaid | Fixed, vendor-styled experience |
| Time to value | One-click Master Setup — live in minutes | Multi-step config, topics & permissions |
A general, good-faith comparison against a typical native/built-in CRM AI agent. Exact capabilities vary by product edition, license and release. AgenTorQ is an independent managed package, not affiliated with or endorsed by any model or CRM vendor.
Connect your email, files and 14+ AI models — chat, search and take real actions. Preview it here, then launch your own in one click.
AgenTorQLIVE DEMOChat with your email & files, run 14+ AI models, take real actions — free to start, no setup.
Experience AgenTorQ in ActionReal Lightning components, not mockups of a vision: talk to the agent on any record, wire any model with your own keys, and build & govern it — all on-platform.
Grounded answers that cite live CRM data — then take a real action inline with an ACT card.
14+ providers, hundreds of models via live fetch. Ping, test latency, set a default — bring your own keys.
Set purpose, wire models, attach knowledge, flip features and lock access — agent, group and user level.
AgenTorQ is built to act on your data without ever betraying it — masked, guarded, governed and logged at every step. Only the masked prompt reaches the model you chose.
Sensitive entities are masked before any prompt leaves your org.
Every read and every write is validated at the gate — regex-free, injection-proof.
Secrets live in an encrypted external vault — never exposed to the model.
Isolated per-user tokens with automatic refresh — your connections, scoped to you.
Runs in USER_MODE — your field-level security and sharing model are enforced.
WorkSpace Profiles and per-user overrides — least privilege, no escalation.
Durable activity log, call logs and per-response safety verdicts — never a black box.
Your data stays in your org; approvals gate every write action.
Drop the managed package into your org. No external infra, no data leaving on a side channel.
Provisions the agent and wires up models, connectors, knowledge and access in a single pass.
Open the console on any record — ask, read files, run actions. Grounded, every time.
AgenTorQ isn't a feature — it's a platform. The complete capability surface, grouped the way you'd actually use it.
A model-agnostic, zero-trust agentic AI platform built natively into Salesforce. It reads live CRM records, understands documents and your knowledge base, takes real actions, and runs on any of 14+ AI providers with your own keys — all on-platform.
Yes. AgenTorQ is an open, model-agnostic alternative to a native, single-vendor CRM AI agent such as Agentforce. Instead of being locked to one model family and metered per conversation, it runs on 14+ providers with your own keys, grounds on your documents and RAG knowledge base, opens up REST and MCP connectors, and adds zero-trust controls like named PII masking and an external credential vault.
OpenAI, Azure OpenAI, Google Gemini, DeepSeek, Mistral, Cerebras, Perplexity, SambaNova, MiniMax, Groq, Grok (xAI), HuggingFace, AI21 and self-hosted Ollama — plus a generic OpenAI-compatible adapter for anything else. Hundreds of models are available via live fetch, switchable per agent and task.
With a zero-trust model: it runs on-platform; named PII is masked before any prompt leaves; every read passes a regex-free SOQL guard and every write a DML guard; credentials live in an encrypted external credential vault with per-user OAuth and token refresh; and access sets control what each agent can read. Only the masked prompt reaches the model you chose.
Yes — create and update records, run Salesforce Flows, send email and show edit forms, with your access rules enforced.
Yes. It has built-in document intelligence for XLSX, PDF and CSV, plus OCR for scanned documents and images.
REST connectors and Model Context Protocol (MCP) connectors with automatic tool discovery, a connector registry and installer, per-user OAuth (Gmail & Google Workspace), automatic Remote Site Settings and full call logging.
An AI-native, VS Code-style IDE built into the console: three panes (explorer, tabbed editor, and a docked Dev Workmate that knows the open file & selection). Browse, edit and deploy Apex, LWC and Flows, with Quick Open (Ctrl/⌘ P), in-file find (Ctrl/⌘ F), an integrated terminal (sf CLI over the org, git over GitHub), a test-coverage gutter, related-test jump, a Flow diagram visualizer, a senior code review, and Apply-with-AI — describe a change or select code, get a diff, approve, then deploy · test · fix.
Yes. DevSpace Studio uses a code-provider registry and GitHub is a first-class provider — browse repos to files, edit and commit, and run real git over the GitHub API (create branch, open PR, list PRs, merge), all through your connector with no local clone.
Yes — a Perplexity-style “Add custom connector” flow: name it, give it an icon and an MCP URL, and sign in once. It uses OAuth 2.1 PKCE with Dynamic Client Registration, discovers the server's tools automatically, and the agent calls them with your approval. Run as many custom MCP servers as you like.
218 agency Workmate templates across 16 divisions (engineering, design, marketing, sales, product, security, finance, support, GIS, game-dev and more). Each ships with a full purpose, instructions and guardrails; one click creates a real, editable Workmate.
Yes. Each Workmate keeps its own memory — Concepts, Entities, Workstreams and Notes — viewable and manageable in a dedicated Memory tab. It applies that memory on future turns, on top of cognee-backed cross-session semantic recall, isolated per user and private by default.
Use AgenTorQ solo, or as a WorkSpace: an admin invites users, builds SFDC-style Profiles from a feature-toggle grid, and assigns them with per-user overrides. Workmates, knowledge base, engine keys and the app catalog are shared; personal chats and OAuth tokens stay private. Enforced operating modes (Ask, Edit, Plan, Auto, Bypass) and deny-by-default, server-enforced access round it out.
Five enforced operating modes you switch per turn: Ask before edits (proposes a diff and waits for approval), Edit automatically (applies safe edits then runs tests), Plan (designs the change and never deploys), Auto and Bypass. The rules are enforced server-side, and destructive actions such as deletes always require confirmation.
Install the managed package and run the one-click Master Setup — it provisions the agent and wires up models, connectors, knowledge and access in a single pass. Live in minutes.
AgenTorQInstall AgenTorQ, run Master Setup, and put a real agentic platform — your models, your data, your rules — to work inside Salesforce.
AgenTorQSign in with Google or Microsoft, or use your email, to open your personal AgenTorQ Workspace — takes seconds.
New here? Entering an email signs you up — we'll email a verification link first. Your sign-in carries straight into your AgenTorQ Workspace.