Platform tour
The complete agentic engineering platform.
Point tools give you an agent. AnyForge adds the planning, delivery, governance and insight layers around your agents, on any model, from frontier clouds to private LLMs on your own infrastructure. From raw customer signal to merged, audited pull request: one platform, one ledger, one flat fee.
Plan
From raw demand to a runnable spec.


Signals: capture demand where it lands
Client asks, sales requests, stakeholder ideas and strategy bets land as Signals. Each keeps its source, so you can tell customer-driven from strategy-driven work.
Opportunities with RICE scoring
Signals become scored Opportunities: Reach × Impact × Confidence ÷ Effort, computed from anchored buckets, with rationale on every input. AI Suggest drafts scores from your signals.
Revenue at Risk
The backlog in money: monthly MRR blocked by unshipped work, annual run-rate exposure, and concentration by region, account and reason.
Roadmap themes with investment guardrails
Now, Next and Beyond horizons with a quarter timeline. Set an investment mix, say 60% features, 25% tech debt and 15% keep-the-lights-on, and see when reality drifts.
Capacity forecasting (Monte Carlo)
Models operator review attention, not engineer hours, as the bottleneck. P50, P85 and P95 Monte Carlo forecasts over your trailing eight weeks, and a what-if simulator.
Spec-driven development (ICE)
Specs and program briefs are hard-gated at plan approval until ICE scores 100%. Simple work can waive optional sections per field; an owner-only override is audit-logged.
- Intent
- Context
- Expectations
Programs → Initiatives → Work items
A three-level hierarchy: cross-codebase programs, per-repo initiatives, and PR-sized work items in dependency waves.
- Bounded or long-lived
- Blocked without losing state
- Cross-initiative dependency map
- Program roll-up, PDF or Markdown
Business Capability Map & Value Chain
A BIZBOK-style capability tree built from your own codebases, with coverage heatmap badges for gaps and duplication. One click projects it onto Porter’s Value Chain.
- Payments
- Risk
- Onboarding
- Export: Markdown, PNG or PDF
Community board: demand from the people who use it
On a public board, users file bugs, features and enhancements with screenshots, upvote them and track them from Received to Shipped.
- Oracle-guided ICE checks
- Anonymized Explore feed
- Delivery pulse
- Each request becomes a work item
Build
Governed delivery that ships real PRs.
Parallel specialist review on every PR
When the PR opens, Architect, Security and QA specialists review the actual diff in parallel: OWASP-class findings, coverage analysis and architectural conformance, attached to the approval card.


One Engineer, on-demand specialists
One Engineer does the execution, with Architect, Security & Compliance and QA specialists as advisors during the run. No handoffs between executing agents.
Deterministic edits, real tests
Patches apply through git and are checked first, never fuzzy-matched, so a stale diff is rejected. Your own test suite is the ground truth.
A fifty-tool agent arsenal
Runs your terminal and searches the live web. For UI changes, it drives localhost and attaches screenshots and console errors to the PR.
- Read-only research sub-agents
- Plan mode
- Isolated git worktrees
- AST and semantic search
- Line-anchored PR reviews
Integrations verified against the real thing
Crews verify real API, payments and webhook integrations in a live sandbox, never holding your provider credentials or egress. A failed check holds Approve behind an explicit override.
Cross-repo context, single-repo blast radius
Sibling repositories mount read-only for cross-repo context. Edits and the PR stay confined to the one writable repo, enforced at the tool layer.
- Shared libraries
- Upstream API producers
- Design systems
- Monorepo migration as governed work
Code Studio: governed coding in the browser
A full agentic workspace on any device, with 200+ models switchable per request. Commands run in an in-browser container or fail over to AnyForge Cloud Agents.
- Files
- Diff
- Terminal
- Metrics
- GitHub repositories
- npm test from your phone
One-click escalation to Crew
When a task outgrows one agent, Code Studio proposes a Crew inline. Accept once and the initiative, work items and crew are created atomically.
Automations & routines
Saved crew templates run on cron schedules or signed webhooks (HMAC-SHA256). Recurring engineering runs itself; you review the PRs.
- Nightly security audits
- Weekly tech-debt refactors
- Dependency sweeps
Playbooks: an operator workflow in one click
Bundled, ready-to-run Operator Workflows that clone into your workspace for customization. Playbooks read production state through an ask-the-Oracle step before they act.
- Production Watchdog
- Cost Anomaly Watch
- Release Health Report
- Security Findings Triage
Self-healing releases
When your pipeline reports a release failure, AnyForge opens a P0 hotfix crew and files the incident. Opt in and it can auto-merge small, safe-path-only hotfix PRs.
- Change-failure rate
- Lead time
Automatic merge-conflict resolution
A sweep every 30 minutes finds cross-work-item conflicts before they block a release. A crew merges (never rebases), runs your tests and updates the PR in place.
Mission control for every run
A live activity feed with a phase tracker, and typed cards for everything that needs you: plan approvals, agent questions, task proposals and ADR revisions.
- Engineer
- Specialist Review
- PR Review
- Desktop alerts at a gate
A review room, not a diff dump
Every crew PR opens in a three-pane review cockpit: an AI walkthrough, syntax-highlighted diffs with click-to-comment, and agent chat to approve, amend or auto-merge.
Workspace bootstrap that respects your repo
Cloud Agents ship every major toolchain, plus cloud CLIs. Pin a setup command, a test command and your AGENTS.md conventions; every crew and Studio session honors them.
- Node
- Python
- Go
- Rust
- Java
- Kotlin
- .NET
- Ruby
- PHP
Code intelligence
The graph index: your codebase as a queryable brain.
A graph of everything you ever shipped
A call-and-import graph of every connected codebase. Agents and people navigate by structure, not by grep.
- TypeScript and JS: compiler API
- Go, Kotlin, Python, Java and Rust: tree-sitter
Semantic search over symbols, not strings
Vector embeddings of symbol-aligned code chunks power natural-language search across every connected repository.
- “Where do we validate webhook signatures?”
Always fresh, never a batch job
Indexing is SHA-gated and incremental. It refreshes on every PR merge via webhook, on operator demand and on a rolling sweep.
Edges that admit what they know
Every call edge carries a confidence label, so agents know when to double-check before acting on it.
- Extracted: import-verified
- Inferred: unique name match
- Ambiguous
Analysis on every pass
Each index run detects god nodes, module communities, cross-boundary surprises and doc-to-code drift, then suggests questions to ask. Incremental runs report the graph delta.
One index, every consumer
The same graph powers the Engineer mid-build, the PR-review specialists, the Oracle, semantic ⌘K search and the Code Graph views, per repo and cross-repo.
Documentation engine
Documentation that regenerates itself.
Deep analysis on connect, fresh on every merge
Point AnyForge at a repository and it reverse-engineers the architecture. Each merged PR patches the affected docs, and every edit is versioned and audited.
- Components
- Boundaries
- Data flows
- Build tooling
- Test posture
C4 diagrams & ADRs, generated
Context, container and component views drawn from the code, plus Architectural Decision Records drafted by the architect agents as they work.
Deployment topology & dependency maps
A live L1 deployment topology with component-to-component dependency arrows, plus cross-repo dependency maps across the whole estate.
Opinionated architecture lenses
Conway analysis asks whether the org chart matches the architecture. The AI-readiness score per codebase and the security-posture view regenerate on merge.
The org narrative
A background worker synthesizes every analyzed system into one business-readable overview, with live generation progress and ETA.
Searchable, shareable, exportable
Full-text search across generated docs, external share links for auditors and diligence teams, and Markdown, PNG and PDF exports.
AnyForge Oracle
The one teammate who has read everything.
A 360° view of every codebase
Sits on the full code-intelligence layer: call graph, semantic search, architecture docs, ADRs, capability map, releases and dependency maps. Answers come with clickable deep links.
Correlated with production reality
Reads your observability stack live and links a production error to the code path, release and PR that introduced it.
- Datadog
- Sentry
- Google Cloud Logging
- Crashlytics
- AWS CloudWatch
- Aikido
Real computation, not vibes
A sandboxed, credential-isolated Python and Node interpreter lets the Oracle parse, aggregate and chart inside the conversation, so answers are computed.
For every member, on every page
Summonable with ⌘-J on every page, for every member including viewers. Read-only, never budget-gated in the console, and it runs on your own AI key.
Answers that leave the chat
Every answer exports to Markdown, JSON, CSV, HTML or DOCX, and ⌘K search matches your generated architecture docs by title and type.
- Per-user memory across sessions
Decision support right at the gate
Consult the Oracle on a HIL approval or halted run opens pinned to its crew, with the gate, verdicts and budget. It advises; it never rubber-stamps.
A confirm-first action tier for operators
Operators can have the Oracle act, with each step confirmed first and role-checked server-side. It becomes a doer only when you say yes.
- Dispatch, amend, halt or resume a crew
- Approve a gate
- Dispatch the architect
- Close a finished run as delivered
- Connect GitHub repos and start analysis
Ask about a person, not just a project
Per-operator answers: what someone launched and delivered, and which gate holds them up. Colleague queries are limited to owners and operators. It reports what happened, not intent.
- “What did I ship last week?”
- “What’s blocking Priya?”
Everywhere your team already is
The same Oracle answers in Slack and from your coding tools, over the headless Control API.
- /anyforge slash command
- @AnyForge mention
- anyforge_ask_oracle MCP tool
- Headless Control API
The agents
Not a bot. A staff.
Opportunity Genesis: from signal to PRD
Drafts versioned PRDs grounded in your signals, with browsable HTML UI mockups your stakeholders can click through before any code exists.
Program Genesis: the portfolio planner
Analyzes your repositories, proposes program scope, creates initiatives, writes versioned program briefs, maintains the risk register and generates status reports.
Spec Genesis: the spec author
Drafts and refines ICE specs in conversation, dispatches the architect for decomposition and requests backlog refinement.
Report Genesis: dashboards you talk into existence
Describe what you want to see and it assembles stat tiles, timeseries, bars and tables, executed live at view time. Export to PDF or Markdown.
- Crews
- Work items
- Initiatives
- Token usage
- Codebases
- Saved under Plan → Reports
The delivery crew & Retro Analyst
After a crew completes, an operator-triggered Retrospective Analyst reads the step trail and audit log, writes the retro and proposes memory facts for people to promote.
- One Engineer
- Architect
- Security & Compliance
- QA
- Sage
Govern
Autonomy you can hand an auditor.


One human gate. Cryptographic proof.
Every normal crew run ends at a mandatory human gate the model cannot bypass. Each decision records a SHA-256 proof of approver, time and content.
- Refused while CI is red, pending or in conflict
- A failing check needs a recorded override
Coverage as a gate, not a vibe
Off by default and fail-open. Turn it on per codebase, and a crew below your PR-head coverage threshold holds the ship.
- Measured coverage, not a self-attested number
Hash-chained audit trail
Governance events, including approvals, constraint checks, crew lifecycle and access, are SHA-256 hashed and chained in an append-only ledger that application code cannot write.
- Tampering breaks verification of everything after it
- Regulator-ready and exportable
Enforced constraints (Atomic Facts)
Your engineering rules are injected into every agent’s prompt and verified per constraint at PR time. Deleting a fact removes enforcement and keeps its audit record.
- TLS requirements
- API contracts
- Schema invariants
- Compliance mandates
Guardrails in code
The crew runtime enforces circuit breakers, and offers each role only the tools on its list.
- Step ceiling
- Loop detector
- Coverage-plateau halt
Portable governance memory
Architecture decisions, API contracts and declined proposals persist across runs, sessions and provider switches. Policy-checked, fail-closed, tenant-isolated, and every write is signed into the audit chain.
Autonomy is a dial, not a switch
Unattended operation is opt-in, one envelope at a time. Auto-merge is opt-in too, within the limits you set. Everything above the envelope waits for a human.
- CI-failure triage
- Auto-fix with capped rounds and attempts
- Auto-merge for small fixes
- Release Health auto-heal on safe paths
- Set at org, program, initiative or gate
Config Packs: compliance in one click
Installable bundles of agent configs, enforced constraints, routines and skills, listed below.
Team & access management
Invitations, multi-domain email auto-join and a role model down to surface-locked board viewers. Service accounts keep working whoever leaves, with scopes capped by a fixed role.
A deliberate trust boundary
AnyForge produces and merges pull requests. It never deploys and never holds your infrastructure keys. Secrets live in Google Secret Manager; the database holds only pointers.
- Commands sandboxed per crew
- Egress proxy refuses cloud-metadata and private addresses
- Per-provider kill switch
Single sign-on, enforced
Connect your identity provider over SAML or OIDC, then turn on Require SSO so it becomes the only way in.
- Owners keep every sign-in method for break-glass
- TOTP two-step on your own account
Sessions and keys that expire
A maximum session length and idle timeout, checked on the server and in the browser. API keys carry an expiry, and a company maximum applies to existing keys.
IP allowlists
Name your networks in IPv4 or IPv6, and requests from anywhere else are refused across the Console, API, MCP and CLI.
- Shared links can be held to the same list
Your keys, or your own tenant
Customer-managed keys from your own Cloud KMS envelope-encrypt your most sensitive data. Revoke the key and nobody at AnyForge can read that data.
- Single-tenant Google Cloud, via Terraform
- CMEK on every data store, in your chosen region
Audit export to your SIEM
Pull the audit trail through a read API with its own audit:read scope, or stream each entry, with its chain fields, to your SIEM.
- Splunk HEC
- Datadog Logs
- Webhook
- Daily chain verifier
- Daily archive nobody can rewrite
Retention, legal hold and deletion
Set how long each record class is kept, from Oracle conversations to crew transcripts. A legal hold stops every deletion; the audit trail is never deleted.
- Owner-requested company deletion, after a cooling-off period
A Trust Center you can check
The public Trust Center shows the security overview and where each compliance programme stands. The SOC 2 programme is in progress; no report has been issued yet.
- NDA requests for security documentation
- Sub-processors published
- Blocking secret scan and dependency audit
- CycloneDX SBOM per release
Config Packs
A governance posture, installed in one click.
PCI-DSS
Cardholder-data guardrails in every agent prompt: encryption, TLS minimums and access logging, plus a weekly routine that checks recent commits.
SOC 2 Type II
The trust-services common criteria as enforced constraints: logical access, change management, monitoring and backup-recovery testing, plus a quarterly access-log audit.
HIPAA Security & Privacy
Security Rule safeguards for ePHI: encryption, unique user identification, audit controls, automatic logoff and contingency-plan testing, with a monthly §-cited report.
HIPAA + SOC 2 (Healthcare SaaS)
One cross-mapped control set for healthcare SaaS: each constraint cites its CFR section and SOC 2 criterion. Install it instead of the two standalone packs.
NIST Cybersecurity Framework 2.0
The six CSF 2.0 functions as engineering constraints: Govern, Identify, Protect, Detect, Respond and Recover, with a quarterly posture review.
NIST SSDF (SP 800-218)
Secure-development practices behind US Executive Order 14028: code integrity, pinned dependencies, pre-merge static review and threat modelling, with prompts that cite practice IDs.
NIST AI RMF 1.0
For crews building AI features: model and system cards, risk mapping, real-metric evaluation, documented human oversight, prompt-injection hardening and a quarterly AI-risk review.
EU Consumer Credit Directive 2
EU CCD2, fully applicable 20 November 2026, for BNPL and consumer-credit software: creditworthiness checks, human intervention on automated decisions and 14-day withdrawal. Engineering guidance, not legal advice.
CNSA 2.0 Post-Quantum Cryptography
The NSA CNSA 2.0 suite as enforced constraints, and a remediation skill that never claims compliance while a non-CNSA algorithm is enabled. Engineering guidance, not a certification.
Software Escrow
Deposits must be buildable and operable by a skilled third party, with a current dependency register and routines for quarterly refresh and annual audit.
Startup
The other direction: loosened rails for velocity. Feature flags preferred over branch protection, tests required on shipped paid paths, a daily PR digest and cheaper default models.
A pack is a starting posture, not a lock: everything it installs is editable and operator-owned, and packs stack. Your company configuration exports and imports as a manifest, so one proven setup can be reused across companies.
See
Every question answered, from board room to terminal.


Management Overview & board packs
A deterministic leadership view of what shipped and was decided, committed versus forecast dates, cost and change-failure rate. One click prints a board pack.
- Variance history
- Weekly brief email
Usage, traces & spend
Month-to-date token spend and projected month-end, by user, initiative, provider, model, repo and product surface. Full LLM trace waterfalls.
- Recent, highest-spend and error lenses
- Filter by Crew, Code Studio or external agent
- Opt-in run-tree tracing
Operator effectiveness scores
A composite 0–100 score per operator covering outcomes, cost efficiency, spec quality, responsiveness and context reuse, with coaching notes.
Two-ledger spend reconciliation
AnyForge’s metered ledger is reconciled daily against each provider’s own usage API, and leakage is surfaced per provider, model and day.
- Anthropic
- OpenAI
- OpenRouter
Platform health & incidents
A live health dashboard with smoke-test heatmaps. Production errors are fingerprint-grouped into signals; accept one and it becomes a KTLO incident a crew can fix.
- 30-second refresh
- Halt taxonomies
Production Watchdog
An opt-in managed agent diffs your observability signals for new errors, spikes and worsening severity, and the Oracle explains each one.
- Digest to inbox, Slack and email
- Run record every pass
- Auto-opens critical KTLO incidents, if you allow it
- Turn on from the Playbooks gallery
Release notes that write themselves
Push a version tag and AnyForge writes redacted, customer-safe release notes for your changelog and timeline.
- New, improved, fixed, scrubbed of secrets
- Announced to Discord
- Shipped requests announce back to the community
Delivery analytics & DORA
Cost per completed crew is the headline number, pivotable by operator, initiative, codebase, program or team.
- Deployment frequency
- Lead time (median and p90)
- Change-failure rate
- MTTR
- First-pass acceptance-criteria rate
- Amend rounds per initiative
- Token forecast vs actual
- Review overrides vs later failures
- GCP cloud spend by service and SKU
Notifications that find you
Four channels: in-app inbox, email, Slack DM and Discord, with an @mention inbox and per-event preferences.
- Approval alerts by severity
- Slack slash commands to ask the Oracle or file a signal
Extend
Plugged into everything you already run.
Every MCP server. Not a curated few.
Any MCP server works, with OAuth, Bearer tokens, API keys or no auth, per user or company-wide. Read tools are open to everyone; mutating tools are operator-gated.
- Streamable HTTP or SSE
- OAuth, auto-refreshed
- Crews get tools with permission
Agents that build to your design system
Agents read a repository’s root DESIGN.md as the visual source of truth, with an org-wide fallback and managed brand assets.
- Machine-readable tokens and guardrails
- Code Studio, Genesis mockups and crews
- Validated on write
Skills marketplace & curated registries
Reusable prompt packs for your conventions, governance rules and output contracts, scoped by toolchain or codebase and composed into every agent.
- Curated registries pinned to immutable commits
- Import SKILL.md files
- Auto-detected from your repos
- Bundled engineering-discipline skills
Drive AnyForge from your editor
The VS Code and Cursor extension puts approvals, crews and the Oracle in your editor. The MCP server adds ~50 tools to any MCP client.
- Claude Code and Cursor
- Scoped read, write and approve keys
GitHub or GitLab, Linear in step
Crews, codebase analysis and merge-request review work on GitLab.com, self-managed GitLab and GitHub. A native two-way Linear sync keeps issues in step.
- Email intake
- Zendesk escalations
- Intercom escalations
A full HTTP control surface
Everything the console does, an API does: repos, crews, initiatives, programs, work items, approvals, coverage, escrow and signals. SDK-compatible LLM endpoints work by changing one base URL.
Coverage & quality ingestion
Pipe the coverage your CI already computes into per-merge program and initiative metrics. Coverage guides scope; your test suite remains the gate.
- Jest
- pytest
- Go
- LCOV
- SonarQube
- Codecov
- Coveralls
Source-code escrow, verified
Deposits include SHA-256 manifests, build-and-deploy runbooks, dependency registers and masked secret scans. A Replicate rehearsal proves each one rebuilds.
- Pinned SFTP to Escode / NCC
- Quarterly fleet sweeps
One-click marketplace
The popular ones. Everything else plugs in the same way.
- GitHub
- Datadog
- Sentry
- Jira
- Confluence
- Linear
- Notion
- Salesforce
- Stripe
- Snowflake
- Cloudflare
- CloudZero
- AWS Cost & Billing
- Google Drive & Calendar
- Microsoft 365
- Firebase
- Aikido
Connectors & integrations
- GitHub
- GitLab
- Linear (two-way sync)
- VS Code & Cursor extension
- Slack
- Microsoft Teams (notifications)
- Zendesk
- Intercom
- Datadog
- Sentry
- New Relic
- Grafana
- CloudWatch
- Aikido Security
- Jira
- Confluence
- Notion
- Salesforce
- Stripe
- Snowflake
- Cloudflare
- CloudZero
- AWS Cost & Billing
- Google Drive
- Google Calendar
- Microsoft 365
- Firebase / GCP
- SonarQube
- Codecov
- Coveralls
- Escode / NCC
- GitHub Actions
- GitLab CI
- Jenkins
- CircleCI
- Azure DevOps
- + any MCP server
Any model
Frontier clouds, open weights, or your own metal.
Every frontier cloud
Anthropic Claude, including Fable 5, Sonnet, Opus and Haiku, plus OpenAI and Google Gemini. Bedrock and Vertex for teams that buy through their cloud, with automatic failover.
Open-weights, first-class
Selectable wherever a frontier model is used: agent roles, Code Studio and routing targets. Reach 200+ models through OpenRouter, or route directly to the vendor.
Custom & private LLM hosting
Run models on infrastructure you control: your own Ollama, a vLLM or llama.cpp cluster, or a private gateway we operate, on-prem, in your VPC or air-gapped.
- Same governance and audit chain
- Same flat platform fee as cloud
- Data residency and regulated industries
Claude Max subscription routing
Paste an OAuth token to route agent traffic through your flat Claude Max subscription, instead of metered API rates. Automatic fallback to an API key.
- Self-pacing throttle against utilization windows
- Extended prompt-cache TTL
Smart routing & caching
Intent-based routing sends each call to the cheapest model that clears the quality bar. Prompt caching bills repeat context at the provider’s cache-read rate, a fraction of the input price, and your Usage page shows what it saved each month.
- Rules, regex or embeddings
- Price recommendations per agent role
Benchmark two models on the same work
Run one work item twice as isolated arms, each with its own run profile. A deterministic recommendation ranks them by outcome, tests, findings and price.
- Advisory: you pick the output
- Neither arm can auto-merge
- Retry a failed arm in one click
Dial reasoning effort, per model
Set low, medium, high, xhigh or max on any agent, and AnyForge translates it for each provider. Only models that reason show the control.
- Anthropic thinking effort
- OpenAI reasoning_effort
- Gemini thinking budgets
- OpenRouter reasoning
- Oracle, Genesis and every Crew role
Budgets that warn, guards that stop
Per-codebase and per-initiative token budgets warn as spend reaches them, without stranding a run mid-change. Runaway guards stop a looping agent, and billing limits stop spend.
- Failed calls never accrue
- One flat platform fee, cloud or self-hosted
Providers
- Anthropic Claude
- OpenAI
- Google Gemini
- AWS Bedrock
- Google Vertex
- OpenRouter (200+ models)
- Ollama (self-hosted)
- vLLM
- llama.cpp
- Private gateway (LiteLLM)
- Claude Max OAuth
Open-weights models
- GLM-4.6
- GLM-4.5 Air
- DeepSeek V3
- DeepSeek R1
- Qwen 2.5 Coder
- Llama 3.3 70B
- Mistral Large
- Grok
Why AnyForge
Compare us to anything. We built it for that.
vs. Coding agents (Claude Code, Cursor, Codex, Copilot)
They are agents; AnyForge is the platform above them. Keep them and govern their calls through Control. One ledger covers your IDE agent, browser agent and background crews.
vs. LLM observability proxies (Langfuse, Portkey, Helicone-class)
Watching spend is not governing it. AnyForge adds enforcement: guards that halt runaway runs, policies that block, and approval gates the model cannot skip.
- Crews, specs, releases and incidents
vs. Agent frameworks (LangChain, CrewAI, AutoGen)
Frameworks give you primitives. You still build the harness: checkpointing, HIL, audit, cost control and tenancy, then maintain it. AnyForge is that harness, already built and governed.
vs. Building in-house
Every team that scales agents ends up building this platform: routing, portable memory, policy engines, hash-chained audit, escrow and capacity science. Install Control in five minutes.
One flat fee. A 10M-token trial. Five minutes.
One platform with one price: a flat per-million-token platform fee, bring-your-own-key, with no seats and no tiers. New accounts start with a free trial, and your first Crew run is on us.