Grep AI
Also known as: Parcha, parcha.ai, Parcha Labs, Inc., Parcha AI
A managed workforce of AI agents for research heavy work that has to be right, from due diligence and business onboarding to underwriting and monitoring. Teams build agents from their own procedures or pick ready made ones, run them as swarms across a list, on a schedule or through the API and MCP, and get every finding with its source. Pro is $200 a month, with a free trial and a pay as you go wallet.
Grep AI, formerly Parcha, builds a managed workforce of AI agents for research heavy work that has to be right: due diligence, business onboarding, underwriting, screening and the monitoring that follows. Parcha Labs started in 2023 with compliance agents for banks and fintechs, opened Grep as a research preview in December 2025, and has since retired the original Parcha product line to focus on Grep. Customers on its site include IG, which uses Grep agents for screening reviews in its UK onboarding, the underwriting team at Shopmonkey, and the venture firms Wisdom Ventures and Initialized Capital.
A team describes the job and shares its process, and Grep builds an agent from it: instructions, context files such as policies and rubrics, skills from a library of more than 300, tools, an input form and an output, which can be a sourced report, a decision, a spreadsheet, a slide deck or structured JSON.
A catalog of more than 200 ready made agents covers work such as business verification, company research briefs, vendor risk, contract review and portfolio monitoring.
Agents run from chat or a form, across a list of up to 1,000 rows, on an hourly to monthly schedule, as monitors that report when something changes, or as swarms, where a lead agent splits the work among specialist agents, combines their findings and sends the exceptions to the team.
Research draws on more than 100 built in data sources, including company registries, sanctions lists, courts and market data, and every claim carries its source. Agents also work inside a customer's own accounts, such as Gmail, Outlook, Notion and Salesforce, through custom APIs and MCP connections, and in web applications that have no API through browsers they sign in to. Grep sends repeated steps to smaller models and judgment to frontier models to hold cost, and a team can set an agent's default model and a spending cap for each run.
Swarm and campaign plans are reviewed and approved before they run. Decision outputs return a probability for each answer, so a team decides where the agent acts and where a person reviews the case. Each new agent comes with evaluation questions that reviewers answer as it runs, and Grep turns their feedback into prompt changes the team accepts or declines.
Role based access, scoped credentials, audit logs of every step, tool call and model choice, a fresh sandbox for each run and SOC 2 Type II compliance are standard, with SSO on team plans, and banks can deploy Grep inside their own perimeter. Developers get a public API, TypeScript and Python SDKs, an MCP server and an A2A agent card. Grep is sold self serve, with a free trial, a Pay As You Go wallet and Pro and Ultra plans at $200 and $500 a month, and team plans through sales.
Vendor details
Canonical URL
https://grep.ai
Category
Agent builder
Subcategory
Managed agent workforce for research heavy work, with custom agents, swarms and a catalog for diligence, onboarding and monitoring
Funding status
Independent. Parcha Labs, Inc., founded in 2023 by AJ Asver and Miguel Ríos, who earlier worked at Brex, and backed by Y Combinator, Kindred Ventures and Initialized.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
More than 100 built in data sources need no login, among them OpenCorporates, OpenSanctions, ComplyAdvantage, GLEIF, Middesk, Crunchbase, PubMed, USPTO and court records. Through Composio, agents also reach more than 1,000 apps in the customer's own accounts, such as Gmail, Calendar, Outlook, Notion and Salesforce, and teams can add custom APIs and MCP connections or sign agents in to web applications that have no API. Work reaches other systems through the REST API, TypeScript and Python SDKs, webhooks, an MCP server for Claude Code, Codex and ChatGPT, A2A, and delivery by email, Telegram and WhatsApp.
In practice
A retail broker's onboarding team is buried in screening alerts, most of them false positives. Grep agents built from the team's own review criteria clear the false positives and pass higher risk cases to analysts with the research attached, which took IG's automated alert review from 10% to 30%.
An underwriting team runs every flagged account through a Grep agent before a credit decision. Business records, litigation, adverse media and ownership come back in one sourced report in minutes instead of hours.
A fintech's risk team has a backlog of businesses waiting for review. A swarm verifies each one, screens sanctions and adverse media, assigns a risk rating with its evidence and sends the flagged cases to the team, and a weekly loop rechecks the approved book.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
13.0 / 14 capabilities · 93%
| Integrations & Tool Calling | Full |
|---|---|
|
Agents work inside the customer's own accounts through more than 1,000 integrations Grep offers with Composio, among them Gmail, Calendar, Outlook, Notion and Salesforce, and a team can add its own custom APIs and MCP connections. In Grep's onboarding example an agent pulls an alert from a Salesforce queue and emails the customer for missing information. The agent builder lists each connection's tools for selection, and credentials are scoped and can be rotated or revoked. SourceGrep AI, grep.ai/blog/introducing-connected-accounts-and-authenticated-browser-access, grep.ai/knowledge-base/create-a-grep-agent and grep.ai/enterprise-controlsread 2026-11-12 |
|
| Workflow Orchestration | Full |
|
A swarm runs one job across a whole list. Its workflow canvas gives each agent an assignment, runs agents in parallel, passes one step's results to the next and combines their findings before export, and a failed record is retried without stopping the rest. Campaign plans in the API declare stages with dependencies, a fan out for each record, joins and exports. SourceGrep AI, grep.ai/knowledge-base/swarms and grep.ai/developers/sdks/campaignsread 2026-11-12 |
|
| Knowledge Grounding & RAG | Partial |
|
Each run gathers its evidence for that run, from more than 100 built in sources such as company registries, sanctions lists, courts and market data, and from the customer's connected accounts, and every claim carries its source. Context files attached to an agent, such as policies, reference tables and rubrics, are staged into the run's sandbox each time it runs. Grep describes no index, graph or embeddings layer kept over the customer's own knowledge across runs. SourceGrep AI, grep.ai/knowledge-base/create-a-grep-agent, grep.ai/developers/mcp and grep.ai/data-sourcesread 2026-11-12 |
|
| Human Oversight & Guardrails | Full |
|
Nothing in a swarm runs until a person reviews the plan Grep drafts and launches it, and the API starts a campaign only on approval of the exact plan reviewed. A swarm brief can limit any action beyond research to none, draft only or propose for approval. Decision outputs return a probability for each answer, and the team sets a threshold above which the agent acts and below which a person reviews the case. SourceGrep AI, grep.ai/knowledge-base/swarms, grep.ai/developers/sdks/campaigns and grep.ai/blog/introducing-decision-outputs-powered-by-jevread 2026-11-12 |
|
| Security, Identity & Governance | Full |
|
Role based access controls who can view, edit and run agents and open their files and reports, and scoped credentials can be rotated or revoked at any time. Team plans add SSO. Grep is SOC 2 Type II compliant with annual third party penetration tests, and its trust center at trust.grep.ai runs on Vanta and shows its detail only in a browser. SourceGrep AI, grep.ai/enterprise-controls, grep.ai/pricing and trust.grep.ai (title only)read 2026-11-12 |
|
| Observability & Auditability | Full |
|
Agent audit logs record every step, tool call and model choice, and the audit trail exports for review. Run history shows jobs, retries, outputs and every file the agent used, a run's work journal lists the browser sessions it opened with a recording of each, and the API returns a timeline for every run. SourceGrep AI, grep.ai/enterprise-controls, grep.ai/changelog and grep.ai/developers/api/v2read 2026-11-12 |
|
| Memory & State Persistence | Partial |
|
Brain keeps a long term memory for each user. An onboarding interview records the user's role, industry, the decisions they make and the people and companies they watch, and later sessions draw on that profile, earlier research and persistent files. Grep states no lifetime for that memory and no way to delete it by itself, and its enterprise controls cover retention and deletion of customer data in general. Credentials saved for browser sign in are a vault for those sessions, not memory. SourceGrep AI, grep.ai/blog/introducing-brain and grep.ai/enterprise-controlsread 2026-11-12 |
|
| Deployment & Data Residency | Full |
|
For banks Grep is managed and deployable inside the customer's own perimeter, and its enterprise offer is a dedicated deployment that keeps data in that perimeter, which Grep's own guide describes as VPC deployment. The self serve harnesses are hosted in the US. Supported configurations and data residency are confirmed in an enterprise review, and no list of regions is published. SourceGrep AI, grep.ai/for-enterprise/global-banks, grep.ai/blog/vpc-deployment-ai-agents-vendor, grep.ai/knowledge-base/create-a-grep-agent and grep.ai/enterprise-controlsread 2026-11-12 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
A public catalog of more than 200 ready made agents covers work such as batch business verification, country risk checks, company research briefs, vendor risk, contract review and portfolio monitoring, and anyone can open one, fill in its form and run it. Each runs on its own, and twenty four of them deliver a finished file such as a spreadsheet or slide deck. More than 300 skills can be added to a team's own agents. SourceGrep AI, grep.ai/agents, grep.ai/for-business and grep.ai/changelogread 2026-11-12 |
|
| Triggers & Channel Coverage | Full |
|
Loops run an agent on a fixed schedule, hourly, daily, weekly or monthly, and monitors check the web at a set frequency and report when new information matching the query appears, with no one starting either run. Schedules can be created, paused and run through the API, and webhooks send job events to the customer's endpoint. Runs also start from chat, a form, the API, MCP and A2A. SourceGrep AI, grep.ai/knowledge-base/monitors, grep.ai/developers/api/v2 and grep.ai/developers/api/webhooksread 2026-11-12 |
|
| Model Flexibility & Routing | Full |
|
In the agent builder's settings a team picks the agent's default model from a list showing each model's relative credit cost, with Opus 5 as the default for new agents, and a swarm's workflow sets the model for each step. The API takes a model for each run, opus, sonnet or haiku, and for every row of a batch or loop. The home page FAQ instead says the harness picks the model for each step against the customer's cost ceiling and accuracy floor. SourceGrep AI, grep.ai/knowledge-base/create-a-grep-agent, grep.ai/knowledge-base/swarms, api.grep.ai/api/v2/openapi.json and grep.airead 2026-11-12 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
The public API v2 is published as an OpenAPI document of 139 operations covering agents, runs, batches, loops, campaigns, files and keys, with scoped API keys, idempotency and webhooks. TypeScript and Python SDKs wrap it, an MCP server with OAuth lets Claude Code, Codex or ChatGPT build agents and run research, and an A2A agent card with its endpoint and auth scheme makes the research agent callable by other agents. SourceGrep AI, grep.ai/developers/api/v2, grep.ai/developers/sdks, grep.ai/developers/mcp and grep.ai/.well-known/agent-card.jsonread 2026-11-12 |
|
| Testing, Debugging & Optimization | Full |
|
When an agent is created, Grep writes a first set of evaluation questions, the team edits them, and reviewers answer them for each job the agent runs. That feedback goes back to the harness, which proposes prompt changes the team reviews and applies one at a time. Grep's guide asks for 10 to 20 test cases scored on correctness, evidence, escalation and schema before an agent is shared. SourceGrep AI, grep.ai/blog/introducing-agent-evals, grep.ai/knowledge-base/create-a-grep-agent and grep.ai/changelogread 2026-11-12 |
|
| Browser & Computer Use | Full |
|
Agents drive browser sessions in Grep's managed environment. A run can open up to four named browsers at once, sign in to web applications that have no API with credentials kept in an encrypted vault, and work through them, with proxies for sites that turn away automated visitors. The run's work journal shows a live view of each session and a recording afterwards. SourceGrep AI, grep.ai/changelog and grep.ai/blog/introducing-connected-accounts-and-authenticated-browser-accessread 2026-11-12 |
|
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
Product updates
Swarms in Grep AI, formerly Parcha, are teams of agents in which a coordinating agent breaks a task into parts, hands them to specialist agents with their own instructions, skills and tools, and combines their findings for review. Uses include merchant categorization, account research, vendor review and customer onboarding with KYB and KYC tasks, run across a list. Agent code runs in isolated environments, and customers control tool and credential access and can set cost limits for each run.
Bears on: Workflow orchestration
View sourcePricing
Pro $200 a month ($167 a month billed annually); Ultra $500 a month ($417 billed annually); Pay As You Go wallet from $20; Free Trial at $0
Credits. A task reserves credits by effort level when it starts and is charged what it actually used. Pro and Ultra include a monthly allowance, and Pay As You Go draws on a prepaid wallet.
Included quota
Pro 1,500 credits a month, about 75 medium effort tasks; Ultra 4,500 a month, about 225; Free Trial 100 one time credits, about 5
What is public
Grep publishes four ways to pay. Pro is $200 a month or $2,000 a year, with 1,500 credits a month, API access, custom agents and enhanced person and company data. Ultra is $500 a month or $5,000 a year, with 4,500 credits a month and premium data included. Pay As You Go is a prepaid wallet from a $20 deposit, with exports, slides and custom agents included and premium data from a $50 balance. The Free Trial gives 100 one time credits. Team plans, with shared agents, pooled credits, SSO and access controls, are quoted through sales, as are dedicated enterprise deployments.
Billing mechanics
A task reserves 2, 20 or 100 credits at low, medium or high effort when it starts, then settles at what it actually used, so a high effort task can cost more than its hold. A completed task costs at least one credit, a failed task is refunded in full, and an API request or an agent can carry a spending cap of $0.50 to $1,000 per run. Wallet balances last 365 days, and subscription credits reset each period.
Overage / add-ons
Pro and Ultra credits reset each billing period and do not roll over. More credits come from a prepaid wallet, which can top up automatically after the first deposit, and a swarm that runs out of credits pauses until more are added.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free Trial at $0 with 100 one time credits, about five medium effort tasks, without exports, custom agents or API access.
Lowest paid plan
Pro at $200 a month, or $167 a month billed annually ($2,000 a year)
Key ambiguities
Grep's pages give different credit figures. The pricing page lists Pay As You Go deposits of $20 for 200 credits, $50 for 500 and $100 for 1,000, while the help center lists $20 for 100 credits, a credit at $0.20 and discounts on larger bundles. The Free Trial carries 100 credits on the pricing page, while the help center and changelog say new personal accounts start with 500 wallet credits.
Related vendors
- Lindy — Lindy is an AI teammate that lives in Slack, answers from your tools…
- Activepieces — Open source, MIT licensed automation platform where every one of 700…
- Agenta — Agenta is an open source workspace where teams build AI agents by…
- AgentX — No code platform to build, evaluate, and deploy multi agent…
- AI Library — No code platform for building and running AI agents, with a…
- Aigensei — Enterprise agentic workflow platform that runs business processes…
Alternatives to Grep AI
The closest documented capability profiles to Grep AI among agent builders tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Agenta12.5 / 14A lighter documented profile than Grep AI
- Gumloop13.5 / 14Fuller documented coverage on Knowledge Grounding & RAG
- StackAI13.5 / 14Fuller documented coverage on Knowledge Grounding & RAG
- Camunda14.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Memory & State Persistence
- Flowable13.0 / 14Fuller documented coverage on Knowledge Grounding & RAG
- FLOWX.AI14.0 / 14Fuller documented coverage on Knowledge Grounding & RAG and Memory & State Persistence
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded