Verdent
Also known as: Verdent AI, Verdent Manager, verdent.ai
Verdent is an agentic coding suite whose desktop Manager turns a goal into a plan of subtasks and runs parallel workers in isolated git worktrees, with reviewer, verifier and planner subagents across Claude, GPT and Gemini models. Work can arrive from Slack, Telegram, Feishu or Discord or run on a schedule, and the same credits cover the VS Code and JetBrains extensions and browser based Verdent Cloud, which publishes web apps.
Verdent, from Verdent AI, Inc., sells agentic coding to founders, solo builders and professional developers who want to hand over a goal rather than manage each step. The desktop app, Verdent Manager, runs on macOS and Windows; extensions bring the same agent into VS Code and JetBrains IDEs; and Verdent Cloud builds and publishes web apps in the browser. The company also runs its own research on coding models, including SWE-bench Verified results.
Manager turns one sentence into stages, subtasks, dependencies and acceptance criteria, dispatches a worker to each subtask and runs as many in parallel as it can, keeping their changes apart in git worktrees and showing progress on a Kanban board. Workers call subagents for fast codebase context, verification, multi model planning and a Reviewer that runs Gemini, Claude and GPT models together and ranks findings by severity. Plan mode settles requirements before building, agent mode asks before destructive or out of scope actions, and finished work waits in a To Review column.
Tasks can be sent from Slack, Telegram, Feishu or Discord and results return to the same thread, Automations run recurring jobs from a plain language schedule, and long term memory, user and project rules, skills, plugins for GitHub, Stripe, Supabase, Notion and Linear, and MCP servers shape what the agent knows and can reach. Models span Anthropic, OpenAI, Google, Moonshot, MiniMax, DeepSeek, Qwen and GLM at provider pricing, with bring your own keys and Claude Code or Codex as alternative worker runtimes.
Lite costs $5 a month for Eco mode on lower cost models, Starter $19 a month with 320 credits, Pro $59 with 1,000 and Max $179 with 3,000, each currently doubled by a promotional bonus, and Teams costs $20 per user a month on Starter terms; Enterprise is custom. Top ups run from 340 credits for $20 to 3,400 for $200 and never expire, and new users get 100 credits for 7 days. Verdent is pursuing SOC 2 and ISO/IEC 42001 certification, and its infrastructure runs on AWS in the United States.
Verdent fits builders and small teams who want an agent that plans, parallelizes, reviews and keeps working from their phone at modest prices. It fits less well for organizations that need a finished SOC 2 report, single sign on, an API to drive the agent, Linux desktops, or hosting outside the United States.
Vendor details
Canonical URL
https://www.verdent.ai
Category
Coding agent
Subcategory
Desktop agent manager with parallel workers, multi model review and chat triggers, plus VS Code and JetBrains extensions and Verdent Cloud app publishing
Funding status
Privately held.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Marketplace plugins for GitHub, Stripe, Supabase, Notion and Linear with OAuth; any MCP server through ~/.verdent/mcp.json; chat channels in Slack, Telegram, Feishu and Discord; models from Anthropic, OpenAI, Google, Moonshot, MiniMax, DeepSeek, Qwen and Zhipu GLM; bring your own keys for Anthropic, OpenAI and OpenRouter; Claude Code and Codex as worker runtimes; web search through Parallel and page fetching through Jina.
In practice
A founder messages Verdent from Telegram to add subscription billing; Manager plans the stages, runs workers in parallel, asks once about the pricing tiers and leaves the finished change with its diffs in To Review.
A developer tags @Reviewer after a refactor; the Reviewer runs three models over the change, lists one P0 and three P1 findings with file and line, and fixes the selected ones.
A growth lead sets an Automation for a weekly report that pulls Google Ads data through a pipeline Verdent built, and the result lands in Slack every Monday.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
11.0 / 14 capabilities · 79%
| Integrations & Tool Calling | Full |
|---|---|
|
Plugins from Verdent's marketplace connect GitHub to view pull requests, manage issues and trigger Actions, Stripe to query payments and manage subscriptions, Supabase to read and write databases and manage users, Notion to query and create pages, and Linear to manage issues and sync status, with OAuth handled in the browser and each package verified before it installs. Any MCP server added to the configuration joins the agent's tool set, and the agent picks MCP tools on its own when they fit. Built in tools cover files, the shell, web search and fetching page content, and Cloud apps get their own database, sign in and analytics. Sourceverdent.ai/docs/verdent-manager/core-features/integrationread 2026-10-18 |
|
| Workflow Orchestration | Full |
|
Manager turns one sentence into a structured plan of stages, subtasks, dependencies and acceptance criteria, decides which subtasks can run in parallel and dispatches a dedicated worker to each, with progress shown live on a Kanban board across projects. Workers call on subagents with their own prompts and models, among them a fast context gatherer, a verifier, a reviewer and a planner that runs Gemini, Claude and GPT models side by side and merges their plans, and custom subagents can be added. Git worktrees keep parallel changes apart, and a person can interrupt mid run while Manager replans. Sourceverdent.ai/docs/verdent-manager/core-features/managerread 2026-10-18 |
|
| Knowledge Grounding & RAG | Full |
|
Manager works from the full project context, and the Fast Context subagent indexes the codebase to locate code and gather evidence quickly. Skills package references, scripts and best practices that load into the agent's context when invoked, the Notion plugin queries docs and knowledge bases, and web search and page fetching bring outside material into a task. Images, design files, mockups and screenshots can be attached as references, so the build follows an existing design or style guide the team supplies. Sourceverdent.ai/docs/verdent-manager/configuration/subagentsread 2026-10-18 |
|
| Human Oversight & Guardrails | Full |
|
Plan mode clarifies requirements, surfaces gaps and risks and produces a plan that can be edited before anyone chooses to start building. In agent mode, file changes outside the expected scope, destructive operations and actions with significant side effects wait for approval. When Manager reaches a fork such as a missing detail, a design choice or a permission, it asks inline and keeps the rest of the work moving, and finished tasks gather in a To Review column with their diffs before anything is accepted. Sourceverdent.ai/docs/verdent-manager/advanced-features/plan-moderead 2026-10-18 |
|
| Security, Identity & Governance | Partial |
|
Verdent is pursuing SOC 2 and ISO/IEC 42001 certification but holds neither yet. Infrastructure runs on AWS in the United States, Vertex and Bedrock model calls carry zero data retention agreements, access to infrastructure is least privilege with multifactor sign in for AWS, model blocklists are respected, and keys for Claude Code or Codex stay on the local machine. Teams brings one bill and a usage view, Enterprise adds organization wide team management, permissions and visibility, and single sign on and audit logs are not offered self serve. Sourceverdent.ai/securityread 2026-10-18 |
|
| Observability & Auditability | Partial |
|
A Kanban board tracks every task across projects with status and a dedicated review column, and each finished task arrives with its plan, outputs, the files it touched, a summary of decisions and a one click path to the diffs or the running app. Automations show each job's last status as done, failed or missed, Teams gives one view of usage, and Cloud apps keep a read only version history and visitor analytics. There is no exportable trace of the agent's tool calls and no hookup to outside monitoring. Sourceverdent.ai/docs/verdent-manager/core-features/managerread 2026-10-18 |
|
| Memory & State Persistence | Full |
|
Manager learns from past conversations, and up to six months of history can be imported so frameworks, naming, architecture decisions and ways of working carry into every new task. Each Manager keeps its own memory and preferences, separate from other Managers, and a duplicated Manager starts with a copy. User rules in VERDENT.md hold personal preferences across all projects, project rules in AGENTS.md hold team conventions and travel through version control, and both are plain Markdown a person can edit. Sourceverdent.ai/docs/verdent-manager/core-features/managerread 2026-10-18 |
|
| Deployment & Data Residency | Partial |
|
Verdent runs as a desktop app for macOS 11 or later and 64 bit Windows 10 and 11, as extensions for VS Code and JetBrains IDEs, and in the browser as Verdent Cloud, with Linux unsupported. The service and its data sit on AWS in the United States only, model calls go through Azure AI Foundry, Vertex and Bedrock, and Cloud projects go live at a verdent.app address with up to 80 GB of storage and a database per app on higher plans. No self hosted service or choice of region is offered. Sourceverdent.ai/docs/verdent-manager/getting-started/supported-operating-systemsread 2026-10-18 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
Built in skills handle Word, PDF, PowerPoint and Excel files, distinctive frontend design and creating new skills, and a Skill Store adds official and community skills that install in one click. Six subagents ship ready to use, among them a fast context gatherer, a verifier, a reviewer and a multi model planner, plugins for GitHub, Stripe, Supabase, Notion and Linear come from a verified marketplace, and Verdent Cloud starts apps from templates. Several Managers can each be set up with their own name, memory and tools and duplicated as starting points. Sourceverdent.ai/docs/verdent-manager/core-features/skillsread 2026-10-18 |
|
| Triggers & Channel Coverage | Full |
|
Work reaches Manager from Slack, Telegram, Feishu or Discord, where each message becomes a tracked task and replies and results return to the same thread, as well as from the desktop app, the VS Code and JetBrains extensions and the browser based Verdent Cloud. Automations turn a request such as a Friday lint and test run or a daily standup summary into a scheduled job, listed with its next run and last status. Manager keeps running when nobody is watching, though scheduled runs depend on the desktop app, and a run missed while the computer was off can be started by hand. Sourceverdent.ai/docs/verdent-manager/core-features/automationread 2026-10-18 |
|
| Model Flexibility & Routing | Full |
|
The lineup spans Anthropic's Claude Fable 5, Opus and Sonnet models, OpenAI's GPT 5.6 family and GPT 5.3 Codex, Google's Gemini 3.6 Flash and 3.1 Pro, Moonshot Kimi, MiniMax, DeepSeek, Qwen and GLM, billed at provider pricing with no markup. Auto mode picks a model per task or a person turns it off and chooses, subagents can each run a different model, and Eco mode uses lower cost models without spending credits. Bring your own keys covers Anthropic, OpenAI and OpenRouter, and workers can run on Claude Code or Codex under the customer's own license. Sourceverdent.ai/docs/account-billing/model-pricingread 2026-10-18 |
|
| APIs / SDKs / MCP Extensibility | Not documented |
|
No API, SDK or MCP server lets an outside tool start Verdent tasks, read their results or manage projects, so the chat bots in Slack, Telegram, Feishu and Discord are the only doors in from outside. Extension runs inward: custom skills, custom subagents written in Markdown, plugins and MCP servers widen what the agent can do, and Claude Code or Codex can serve as the runtime for workers. Sourceverdent.ai/docs/verdent-manager/configuration/byoaread 2026-10-18 |
|
| Testing, Debugging & Optimization | Full |
|
A code review subagent runs Gemini, Claude and GPT models in parallel over a change and returns findings ranked P0 to P2, each with an explanation, a file and line and a rating of how sure it is, and selected findings are fixed automatically. The agent can call that subagent as a final verify step on its own, a lighter Verifier catches obvious issues, and in VS Code a verifier subagent with browser tools automates test interactions. Automations can run linting and tests on a schedule, and a Cloud project needs a working preview before it can go live. Sourceverdent.ai/docs/verdent-manager/advanced-features/code-reviewread 2026-10-18 |
|
| Browser / Computer-use | Partial |
|
In the VS Code extension a verifier subagent uses browser tools to automate test interactions with the app being built, and Design mode and attached screenshots guide what the interface should look like. Web search and page fetching run through Parallel and Jina, which return page content to the agent without it steering a browser. No agent operates other websites, a desktop or outside software on its own, and the desktop app offers no computer use, so the browser work stays confined to checking the project in progress. Sourceverdent.ai/docs/verdent-for-vscode/getting-started/overviewread 2026-10-18 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Lite $5 a month; Starter $19; Pro $59; Max $179; Teams $20 per user
Monthly subscription with a credit allotment for Prime mode models and Eco mode capacity, plus never expiring top ups; Teams per user
Included quota
Lite: Eco mode light usage, no credits. Starter: 320 credits (480 with the current 50% bonus). Pro: 1,000 (1,500). Max: 3,000 (4,500). Teams: Starter terms per user. Deployed apps 2, 2, 10 and 50.
What is public
All plan prices, base and bonus credits, top up packs, the per credit rate and deployment limits.
Billing mechanics
Monthly Stripe subscription with credits refreshing each cycle, one shared pool across Desktop and VS Code, and never expiring top ups.
Cost watchouts
The 50% credit bonus and the deployment limits are promotional and may change; parallel workers on frontier models consume credits quickly.
Variable cost rationale
Credits track provider token prices, so frontier models and large parallel runs drain the allotment fast and need top ups, while Eco mode work costs nothing extra.
Additional watchouts
Promotional bonuses inflate the credit counts shown.
Overage / add-ons
Top ups at about $0.059 a credit: 340 for $20, 1,020 for $60, 1,360 for $80, 1,700 for $100, 3,400 for $200, never expiring; credits are deducted at provider model pricing with no markup; Eco mode continues without credits.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free trial with 100 credits for 7 days
Lowest paid plan
Lite at $5 a month: Eco mode on lower cost models for light usage, credits added as needed, 2 deployed apps; Starter at $19 a month adds 320 credits for Prime mode models
Commercial notes
Teams gives one bill with members added or removed at any time; Enterprise tailors credits, billing and invoicing.
Key ambiguities
How much Eco mode usage each plan allows is described only as light, standard, extended or maximum.
Cancellation / refund
Upgrades are charged in full immediately and reset the cycle, keeping existing credits.
Support SLA / resale
Priority support and onboarding on Enterprise.
Missing data
Annual pricing, Eco mode limits and Enterprise pricing.