ForgeCode
Also known as: forgecode.dev, ForgeCode software factory, Tailcall, Inc.
ForgeCode, from Tailcall, Inc., sells a software factory: cloud machines where coding agents take work from Slack, GitHub issues, forum posts, production alerts and schedules, split each goal into parallel issues, write and test code, pass it through a reviewer agent and security gates, and open or merge pull requests under a per repository policy. Customers bring keys for more than 300 model providers, pricing starts at $0.20 per agent hour, and Enterprise can run factories in the customer's own cloud.
ForgeCode is a software factory sold by Tailcall, Inc., a Delaware corporation, with Tushar Mathur as founder. Each factory is an isolated cloud machine where coding agents take engineering work from wherever it starts, carry it through code, tests and review, ship it, and keep watching production afterward. The terms describe the service as in early access, where access may need an invitation, and the company names no customers or funding on its own pages.
Work arrives from Slack mentions of @ForgeCode, GitHub issues, Discourse forum posts, Sentry alerts, Linear epics, schedules and plain prompts, and watchers nudge or escalate when an issue sits idle or a pull request waits. A mission board splits each goal into issues with priorities and dependencies, and each issue runs in parallel on its own git worktree and branch.
Changes pass the repository's tests and linters, security scans and an adversarial review agent that did not write the code, and a merge policy set per repository either waits for a person or merges once the gates pass. Factories checkpoint as they go and restart on their own from the last checkpoint after a crash, post results back to Slack, send a weekly executive summary, and keep skills and memories about the codebase and its incidents that they refine over time. ForgeCode presents twelve engineering roles a factory can take on, from on call and bug triage to migrations, releases and accessibility.
Pay as you go costs $0.20 per agent hour on a small machine, rising to $1.60 on the largest, with each agent billed separately, and models are paid either through the customer's own keys for more than 300 providers, including Anthropic, OpenAI, Vertex AI and DeepSeek, or through ForgeCode's gateway at API prices from prepaid credit. A launch offer turns $1 into $100 of credit for 24 hours.
Enterprise is priced case by case and adds factories in the customer's AWS, Google Cloud or Azure account, organizations and teams, SAML single sign on and a forward deployed engineer. Managed factories run on Google Cloud in the central United States, customer content is not used for training, and security reports sit behind a request form on its trust center.
It fits engineering teams that want agents running long programs of work, such as bug triage, dependency upkeep, migrations and on call fixes, with merges gated by policy and reports for engineering leaders. It fits less well for buyers who need a public API to drive it, a published attestation before signing, or team roles and permissions below the Enterprise plan.
Vendor details
Canonical URL
https://forgecode.dev
Category
Coding agent
Subcategory
Software factory of cloud coding agents that take work from Slack, GitHub, alerts and schedules and ship pull requests under merge policies
Funding status
Private company, Tailcall, Inc. of Delaware, selling as ForgeCode, with Tushar Mathur as founder. It names no funding or investors.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Slack app that answers @ForgeCode mentions and posts results and weekly summaries, GitHub for issues and for opening and merging pull requests, and work taken from Sentry alerts, Discourse forum posts and Linear epics. Agents search the web through Exa and code through Voyage AI embeddings, and models run on keys for more than 300 providers including Anthropic, OpenAI, Vertex AI, DeepSeek, OpenRouter, Fireworks AI and Z.ai.
In practice
An engineer mentions @ForgeCode in a Slack thread about a flaky checkout test. The factory opens an issue on its mission board, reproduces the failure on its own branch, fixes it, passes tests, linters and the review agent, and holds the pull request for approval because the repository's merge policy is ask me.
A Sentry alert fires overnight. The factory picks it up, traces the error, opens a hotfix pull request and merges it once the gates pass, because the policy for hotfix branches is merge when gates pass, then posts the result to the incident channel.
A platform lead gives the factory a migration goal on Monday. The mission board splits it into dependent issues that run in parallel, and the weekly summary to engineering leads reports throughput, incidents and what is left.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool Calling | Full |
|---|---|
|
ForgeCode runs as a Slack app that answers @ForgeCode mentions in the same thread, posts results and sends a weekly summary, and in GitHub it replies on issues and opens and merges pull requests once the customer connects it. Its examples take work from Sentry alerts, Discourse forum posts and Linear epics, the agents search the web through Exa, and code search runs on Voyage AI embeddings of the customer's repositories. Sourceforgecode.devread 2026-11-08 |
|
| Workflow Orchestration | Full |
|
A mission board in each factory splits a goal into issues with priorities, runs and dependencies, and each issue runs in parallel as its own conversation on its own git worktree and branch, worked by a fleet of agents under a mission planner. An adversarial review agent that did not write the change checks it before a pull request opens, issues can block one another, and a factory checkpoints its work and restarts from the last checkpoint on its own after a crash. Sourceforgecode.dev/blog/what-is-a-software-factoryread 2026-11-08 |
|
| Knowledge Grounding & RAG | Full |
|
When semantic code search is turned on, ForgeCode stores code chunks and their embeddings from Voyage AI on its servers, so agents query an index of the codebase that persists between runs, and conversation sync can keep past conversations alongside it. Agents also search the web through Exa, load skills written into the repository as procedures, and read the current code directly on each issue's own worktree. Sourceforgecode.dev/privacyread 2026-11-08 |
|
| Human Oversight & Guardrails | Full |
|
Each repository carries a merge policy: ask me, which holds a pull request until a person approves it, or merge when gates pass, which can be set for a branch pattern such as hotfixes. A reviewer agent fixes findings before a pull request opens, watchers send reminders when a pull request waits, and teams widen autonomy one class of change at a time, with people approving the spec, reviewing the code and judging the result. Sourceforgecode.devread 2026-11-08 |
|
| Security, Identity & Governance | Partial |
|
Enterprise adds SAML single sign on and organizations and teams, while the console signs in with email and password, GitHub or Google through an OpenID Connect realm. Data is encrypted in transit and at rest, cloud machines are isolated per user, staff access is least privilege with multifactor sign in, secrets sit in Google Cloud Secret Manager, and customer content is not used to train models. ForgeCode names no SOC 2 or ISO attestation, and its trust center opens only through a request form. Sourceforgecode.dev/privacyread 2026-11-08 |
|
| Observability & Auditability | Partial |
|
A mission board shows each issue's runs and dependencies, factory status shows whether each factory is online, paused or offline with its uptime, and runs keep execution logs and crash logs. Live dashboards and a weekly executive summary report throughput, quality, incidents and where work came from, and artifacts include test reports and video walkthroughs of changes. No audit log of who approved or changed what is offered. Sourceforgecode.dev/pricingread 2026-11-08 |
|
| Memory & State Persistence | Partial |
|
Each factory keeps knowledge, which ForgeCode describes as what it remembers about the customer's codebase and incidents, limited on pay as you go and set by contract on Enterprise, and it refines those memories along with its procedural skills over time. Skills are written to the repository as procedures the agents search and load later. No lifetime, review screen or delete path for the factory's memories is given. Sourceforgecode.dev/pricingread 2026-11-08 |
|
| Deployment & Data Residency | Full |
|
Managed factories run on Google Cloud in the central United States, and Enterprise customers can run factories in their own AWS, Google Cloud or Azure account instead, keeping code and execution inside infrastructure they control. Machine sizes run from 2 vCPU and 8 GB of memory to 16 vCPU and 64 GB, and model calls can run on the customer's own provider keys rather than through ForgeCode's gateway. Sourceforgecode.dev/pricingread 2026-11-08 |
|
| Prebuilt Agents / Templates / Packs | Partial |
|
ForgeCode presents twelve engineering roles a factory can take on: on call, bug triage, spec to feature, migrations, QA and tests, CI and dependencies, security fixes, releases, performance, docs and SDKs, tech debt and accessibility. On Enterprise a forward deployed engineer sets a factory up for the chosen use, and there is no library of packaged agents or templates for customers to install themselves. Sourceforgecode.devread 2026-11-08 |
|
| Triggers & Channel Coverage | Full |
|
Work starts from Slack mentions, GitHub issues, Discourse forum posts, production alerts from Sentry, Linear epics, schedules set as crons and plain prompts. Automations act as triggers and watchers that start, nudge, rerun and escalate work, for example when an issue has sat idle for six hours or a pull request has waited eight, and a weekly summary goes out to an engineering leadership channel on its own. Sourceforgecode.dev/pricingread 2026-11-08 |
|
| Model Flexibility & Routing | Full |
|
Customers bring their own keys for more than 300 model providers on every plan, including Anthropic, OpenAI, Vertex AI, DeepSeek, OpenRouter, Fireworks AI and Z.ai, and pick the model and mode a factory uses, with provider charges billed by the provider. ForgeCode's own gateway offers a few tuned models with no keys needed, billed at API prices from prepaid credit. Sourceforgecode.dev/pricingread 2026-11-08 |
|
| APIs / SDKs / MCP Extensibility | Not documented |
|
ForgeCode offers no public API, SDK or MCP server for its factories, and its terms bar using the service's API outside ForgeCode's own applications or as a backend for another product. Work reaches a factory through Slack, GitHub and the other connected channels, and the console is where customers manage it, so outside systems drive a factory only through those channels. Sourceforgecode.dev/termsread 2026-11-08 |
|
| Testing, Debugging & Optimization | Partial |
|
Every change passes gates before merge: the repository's own tests and linters, security scans, staging smoke tests, load tests and the adversarial reviewer, shown as a verification count per change. A blog post describes the factory keeping a change to its own skills only when it beats the best result so far, and each shipped change records value, cycle time, defects, tokens and compute. The gates run on the customer's code with the customer's own tests. Sourceforgecode.dev/blog/what-is-a-software-factoryread 2026-11-08 |
|
| Browser / Computer-use | Not documented |
|
Artifacts include staged demos and video walkthroughs of changes, and the accessibility role runs axe audits, which point to factories running the apps they build in an interface. ForgeCode does not say whether its agents drive a browser or a desktop, and its documentation site refuses automated reads, so how those walkthroughs are made stays open. Sourceforgecode.dev/pricingread 2026-11-08 |
|
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
Pay as you go from $0.20 per agent hour on a Small machine, with each agent billed separately and models paid through the customer's own keys or ForgeCode's gateway.
Agent hours by machine size, billed per agent for the hours actually used, plus model usage billed by the customer's provider or at API prices through ForgeCode's gateway.
Included quota
Pay as you go includes limited automations, artifacts, knowledge, reports, crons and cloud machines, with no numbers stated; Enterprise sets custom limits.
What is public
Per agent hour rates for four machine sizes, the bring your own key and gateway model billing, and the launch offer.
Billing mechanics
Prepaid credit and metered agent hours per machine size, with model usage billed by the customer's provider or drawn from credit at API prices through the gateway.
Cost watchouts
Agents bill separately, so parallel issues multiply agent hours, model usage bills on top of agent hours, the launch credit expires within 24 hours, and the pay as you go limits on automations, knowledge and machines carry no stated numbers.
Variable cost rationale
Everything is metered: each agent bills by the hour at its machine size, factories run around the clock on triggers, schedules and watchers, and model tokens bill on top through the customer's provider or the gateway. A factory running several agents in parallel on larger machines multiplies the hourly rate.
Additional watchouts
Plan limits are labeled Limited without figures.
Overage / add-ons
Every agent hour is billed at its machine's rate: $0.20 Small, $0.40 Medium, $0.80 Large and $1.60 Ultra, so three agents running an hour on Small cost $0.60. Gateway model use draws on prepaid credit at API prices.
Sales call required
Mixed (some tiers require a call)
Free / trial
There is no free tier. A launch offer gives $100 of promotional credit for $1 of prepaid credit, expiring 24 hours after it is granted.
Lowest paid plan
Pay as you go at $0.20 per agent hour on a Small machine
Commercial notes
Tailcall, Inc. sells ForgeCode self serve and through Enterprise contracts, with the service described as early access in its terms.
Key ambiguities
The pricing page does not state the limits behind its Limited labels, and no page says whether prepaid credit pays for agent hours or only for gateway model use.
Cancellation / refund
Pay as you go with prepaid credit; Enterprise terms priced case by case.
Support SLA / resale
Community support on both plans, and Enterprise adds a forward deployed engineer who sets up the factory.
Missing data
Enterprise pricing and the numeric limits on pay as you go.
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…