Potpie AI
Also known as: Potpie, Potpie AI, potpie.ai, potpie-ai/potpie, Context Engine, Context Graph
Open source context layer that turns a codebase and its development lifecycle into a property graph for Claude Code, Codex, Cursor and OpenCode, sold to large engineering teams as an embedded engagement with forward deployed engineers that runs in the customer's VPC or on premises.
Potpie turns a codebase and its surrounding software development lifecycle into a living context graph for AI agents. It parses a repository into a property graph capturing every file, function, class, import and call relationship, then layers in source history, decisions, tickets and team knowledge. An agent can then answer questions, plan changes, debug failures and write code with context specific to the project. The company was founded in late 2023 by Aditi Kothari and Dhiren Mathur and spent nearly two years building that layer before its public launch.
The architecture now starts from the CLI. Potpie supplies context to the coding agents a team already uses: a setup wizard provisions local config, storage, a daemon, a default workspace and agent skills, then installs Potpie's instructions and skills into Claude Code, OpenAI Codex, Cursor or OpenCode. The harness runs the task, and Potpie tells it what it needs to know first. The CLI is designed to be driven by both humans and agents.
Its commands register sources, resolve the context an agent should read before a task, search for a specific file, workflow, bug, decision or convention, record a durable project learning, and work directly against the graph. A local command opens a graph explorer in the browser. Potpie also offers prebuilt agents for codebase Q&A, code generation, spec generation and debugging, custom agents defined by role, goal and tools, and an API for parsing a repository and holding conversations with an agent.
The context graph is the product. It spans four integrations covering source control, ticketing and documentation: GitHub for repositories, pull requests, issues, reviews and source history, Linear for teams, issues, projects and documents, Jira for projects, issues, status and changelog context, and Confluence for spaces, pages, runbooks and decisions.
Potpie is built for large systems, codebases from around one million lines to hundreds of millions, where the hard part is reasoning across services, dependencies and production signals. It targets high risk work such as root cause analysis and blast radius detection. One customer with a codebase above 40 million lines cut root cause analysis on production issues from nearly a week to 30 minutes.
Alongside the graph sits spec driven development, a workflow that settles requirements and architecture before code is written, so the plan is reviewable ahead of implementation. The Context Engine includes durable memory. Learnings and decisions are written explicitly and retrieved later, so they persist across sessions rather than being rebuilt each time.
The project is Apache 2.0 open source with roughly 5,600 stars, distributed on PyPI, with the daemon, storage and graph explorer all running on the developer's own machine. The commercial offer is an embedded engagement with forward deployed engineers, priced by the project rather than per seat and run in the customer's VPC or on premises.
Its Recipe workflow layer starts work on schedules, repository changes, incident alerts and deployment events. Potpie's trust center lists SOC 2 Type II, ISO 27001 and ISO 42001, while SSO, role based access and audit logging are not documented. Its customers include Fortune 500 and publicly listed companies in regulated industries such as healthcare and insurance technology.
Potpie raised a $2.2M pre seed round in early 2026 led by Emergent Ventures, with All In Capital, DeVC and Point One Capital participating. It also publishes its SWE-bench predictions, execution logs, trajectories and evaluation results as open source, along with research on agent evaluation data and sandboxing.
Vendor details
Canonical URL
https://potpie.ai
Category
Coding agent
Subcategory
Codebase context graph for AI-native SDLC
Funding status
Independent. Founded in late 2023 by Aditi Kothari, chief executive, and Dhiren Mathur, Potpie raised a 2.2 million dollar pre seed round in early 2026 led by Emergent Ventures, with participation from All In Capital, DeVC, and Point One Capital. The company reports Fortune 500 and publicly listed customers in regulated industries and more than five thousand stars on its open source project. Total disclosed funding is 2.2 million dollars.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Four integrations span source control, ticketing and documentation: GitHub (repositories, pull requests, issues, reviews, source history), Linear (teams, issues, projects, documents), Jira (projects, issues, status, changelog context) and Confluence (spaces, pages, runbooks, decisions), connected through CLI auth commands with credential verification. Potpie installs its instructions and skills into four coding harnesses, Claude Code, OpenAI Codex, Cursor and OpenCode, so those agents call it for context during a task. The CLI is designed to be driven by humans and agents alike, exposing context resolution, search, durable recording and graph operations. There is no Notion, Slack or VS Code extension integration.
In practice
Your codebase exceeds a million lines and changes are risky. Potpie builds an ontology first knowledge graph so agents can trace dependencies, run root cause analysis, and detect the blast radius of a change.
You want to standardize on your own models or keep code private. Self hosters set the model provider and the chat and inference models through environment variables, with a local Ollama model as the worked example, and because Potpie is open source you can self host the whole platform.
A new engineer needs to ramp on an unfamiliar service. They ask Potpie's agents how authentication works or how an order flows, and get answers with references drawn from the codebase graph.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
11.0 / 14 capabilities · 79%
| Integrations & Tool Calling | Full |
|---|---|
|
There are four integrations across source control, ticketing and documentation. GitHub covers repositories, pull requests, issues, reviews and source history, Linear covers teams, issues, projects and documents, Jira covers projects, issues, status and changelog context, and Confluence covers spaces, pages, runbooks and decisions, each connected through CLI auth commands with credential verification via potpie auth status --verify. Potpie also installs its instructions and skills into four coding harnesses, Claude Code, OpenAI Codex, Cursor and OpenCode, so those agents call it for context. There is no Notion, Slack or VS Code extension integration. Sourcegithub.com/potpie-ai/potpieread 2026-08-30 |
|
| Workflow Orchestration | Full |
|
Potpie either operates as a single agent or delegates execution to specialized subagents, and customers define custom agents by role, goal, backstory and system prompt with tasks, tools and expected outputs, so it supports both runs across several agents and a flow the customer configures. Sourcedocs.potpie.ai/custom-agents/introductionread 2026-09-29 |
|
| Knowledge Grounding & RAG | Full |
|
The repository is parsed into a property graph capturing every file, function, class, import and call relationship, held in a graph database, then layered with source history, decisions, tickets and team knowledge so agents can perform multi-hop reasoning across components. Integrations index GitHub repositories, pull requests, issues, reviews and source history, Linear teams, issues, projects and documents, Jira projects, issues, status and changelog context, and Confluence spaces, pages, runbooks and decisions. The CLI exposes potpie resolve to pull the context an agent should read before a task and potpie search to look up a specific file, workflow, bug, decision or convention, and Potpie reports operation on codebases from roughly one million to hundreds of millions of lines. Sourcedocs.potpie.ai/introduction and github.com/potpie-ai/potpieread 2026-08-30 |
|
| Human Oversight & Guardrails | Partial |
|
Spec-driven development is the main workflow, putting upfront planning first to define clear requirements and architecture before code is written, so the plan is reviewable ahead of implementation. The CLI surfaces context for review through potpie resolve before a task begins, the graph explorer makes the context inspectable, and potpie graph exposes proposals and commits as distinct operations. There is no per-action approval gate, tool permission scoping or configurable runtime guardrail, and because Potpie supplies context rather than executing changes itself, enforcement of any write path rests with the configured coding harness. Sourcedocs.potpie.ai/introduction and github.com/potpie-ai/potpieread 2026-08-30 |
|
| Security, Identity & Governance | Partial |
|
The trust center at trust.potpie.ai lists SOC 2 Type II, ISO 27001:2022 and ISO 42001:2023, with an on premises option and Google Cloud hosting. SSO, SAML, SCIM and role based access are not documented. Sourcetrust.potpie.ai and docs.potpie.airead 2026-10-25 |
|
| Observability & Auditability | Partial |
|
The CLI reports on the context layer (status, doctor, auth checks, graph reads), and a local graph explorer makes the context inspectable. The homepage states that every run is logged with agent, action, tokens and outcome in an agent activity log. The documentation does not describe that log, how long it is kept or how to export it. Sourcepotpie.ai, potpie.ai/llms.txt and github.com/potpie-ai/potpieread 2026-10-25 |
|
| Memory & State Persistence | Full |
|
The Context Engine includes a graph-backed memory layer for entities, claims, source refs and timelines. Agents propose memory through semantic mutation, the record tool saves durable project memory such as decisions, observations and validated graph facts, and resolve and search read it back, all scoped to one pot. The memory has a stated scope and a durable lifetime, and deletion is not addressed. It is separate from the code graph that grounds knowledge. Sourcedocs.potpie.ai/concepts/context-engineread 2026-09-29 |
|
| Deployment & Data Residency | Full |
|
The platform is published under Apache 2.0 with a CLI-first architecture installed from PyPI, where the setup wizard provisions local config, local storage, a local daemon and a default pot, and potpie ui serves a graph explorer from that local daemon in the developer's browser. The whole platform can be self-hosted, and account-backed managed features are opt-in through potpie login rather than required. The repository includes architecture notes for self-deployment, and Potpie reports deployments in regulated industries including healthcare and insurance technology. Sourcegithub.com/potpie-ai/potpieread 2026-08-30 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Named prebuilt agents each do their own job and are callable through the API, namely a codebase Q&A agent, a code generation agent, a spec generation agent and a debugging agent, beside skills installed into Claude Code, Codex, Cursor and OpenCode. Sourcedocs.potpie.ai/agents/api-accessread 2026-09-29 |
|
| Triggers & Channel Coverage | Full |
|
Recipe, an enterprise workflow layer, runs durable workflows on cron or calendar schedules and on repository changes, incident alerts and deployment events, so work starts without a person invoking it. The open source CLI is invoked by hand or from inside Claude Code, Codex, Cursor and OpenCode. Sourcedocs.potpie.ai/build-flow/reciperead 2026-10-25 |
|
| Model Flexibility & Routing | Full |
|
Self hosters set the provider and models themselves through environment variables (LLM_PROVIDER, CHAT_MODEL, INFERENCE_MODEL), with a local Ollama model given as the worked example, so the customer chooses the model. Sourcedocs.potpie.ai/self-hosting/setupread 2026-09-29 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
An API (x-api-key header, /api/v2 on the Potpie server) parses a repository, creates a conversation bound to an agent and sends messages with streaming and citations, beside the PyPI CLI driven by people and agents and the Apache 2.0 source. Sourcedocs.potpie.ai/agents/api-accessread 2026-09-29 |
|
| Testing, Debugging & Optimization | Partial |
|
Debugging and root cause analysis are example uses of the context layer, with multi-hop reasoning across components, and one customer cut root cause analysis on a 40 million line codebase from nearly a week to 30 minutes. Tests are aligned against a plan before code is written through spec-driven development, and the CLI exposes graph quality checks. Potpie also open-sources SWE-bench predictions, execution logs, trajectories and evaluation results from its own benchmarking, and it publishes research on agent evaluation data. Customers get no evaluation, regression testing or debugging product for agent behavior. Sourcedocs.potpie.ai/introduction, github.com/potpie-ai and potpie.ai/blogread 2026-08-30 |
|
| Browser & Computer Use | Not documented |
|
All operation runs through programmatic interfaces, namely a Python CLI installed from PyPI, a local daemon serving a graph explorer, repository parsing into a graph database, integration APIs for GitHub, Linear, Jira and Confluence, and skill files installed into coding harnesses that those harnesses read. There is no browser control, screenshot capture, visual verification or way to operate software lacking a programmatic interface. Sourcegithub.com/potpie-ai/potpie and docs.potpie.ai/introductionread 2026-08-30 |
|
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
Free and open source to self host (Apache 2.0). The commercial offer is an embedded engagement priced by the project, not per seat, through a demo.
Per project engagement, not per seat
Included quota
The open source edition (Apache 2.0) is the full context layer to self host, with the user supplying model keys and infrastructure. The commercial engagement adds enterprise features such as Recipe workflows, deployment in the customer's VPC or on premises, and forward deployed engineers, priced per project.
What is public
The free open source edition and the pricing basis are public: Potpie says it does not sell AI seats and charges for an embedded engagement by the project. No figures are published.
Billing mechanics
Free to self host as open source, with model inference billed by the provider under the user's own keys. The commercial offer is an embedded engagement priced per project, arranged after a demo.
Cost watchouts
When self hosting, the real recurring cost is model inference. Potpie routes to providers like OpenAI, Anthropic, and Gemini using the customer's own keys, so token spend scales with agent activity on large codebases, which can be significant at millions of lines. Running the graph database and daemon also consumes infrastructure. Enterprise deployments are quoted individually.
Variable cost rationale
Potpie is free to self host, and its agents call external model providers on the customer's own keys, so inference is metered by the provider and grows with usage and codebase size. The commercial engagement is priced per project by contract. No Potpie cap on provider spend is documented.
Additional watchouts
No paid price is published for the managed or enterprise tiers. Self hosting shifts cost to model inference and infrastructure, which scales with codebase size and agent activity.
Overage / add-ons
No Potpie usage meter is documented for the open source edition, and model inference is billed by the chosen provider under the user's own keys. Engagement terms are not public.
Sales call required
Yes, required for paid access
Free / trial
Free open source edition to self host. No trial.
Lowest paid plan
Not published
Commercial notes
Potpie's commercial model splits a free open source edition, aimed at developers and teams that self host with their own model keys, from an embedded engagement with forward deployed engineers for large organizations, priced by the project. On the open source path, the lever that matters most economically is model inference spend on large codebases, since the platform routes to external providers.
Key ambiguities
Engagement prices are not published, so the cost above the free open source edition is not public. For self hosting, the dominant variable is model inference spend, which depends on usage and codebase size.
Cancellation / refund
The open source edition carries no contract. Managed and enterprise terms are arranged directly with the company and are not published.
Missing data
Engagement pricing, scope and minimums are undisclosed. Only the free open source edition and the pricing basis are public.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose AI agents build, host and manage…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Amp — Amp is a coding agent from Amp Frontier Corporation that runs each…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
Alternatives to Potpie AI
The closest documented capability profiles to Potpie AI among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Qodo12.0 / 14Fuller documented coverage on Security, Identity & Governance and Testing, Debugging & Optimization
- Embedder11.5 / 14Fuller documented coverage on Human Oversight & Guardrails and Testing, Debugging & Optimization
- Refact.ai12.5 / 14Adds documented Browser & Computer Use
- Augment Code12.0 / 14Fuller documented coverage on Human Oversight & Guardrails and Security, Identity & Governance
- CodeGPT10.0 / 14Fuller documented coverage on Human Oversight & Guardrails
- CodeRabbit12.0 / 14Fuller documented coverage on Human Oversight & Guardrails and Security, Identity & Governance
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded