The Developer Platform for AI Agents.

Stop building one-off agent tools. Aeion Forge is a production-grade AI developer workspace providing 160+ MCP tools for project management, code intelligence, semantic search, sprint automation, and agent memory synchronization—built natively into your OS.

The AI-Native Developer Stack

160+ MCP Actions

Comprehensive Model Context Protocol toolset — one Forge tool exposes 160+ typed actions covering projects, issues, sprints, specs, decisions, knowledge, agent memory, files, code intelligence, bug capture, and PR tracking. Every action Zod-schema-validated. `describe_action` lets agents query the exact param shape of any action before calling it — eliminates schema guessing.

Code Intelligence Engine

Tree-sitter WASM parses your codebase to extract symbols, signatures, and call graphs from TypeScript/JavaScript/Python/Go/Rust. AI-powered issue breakdown turns a story into sub-tasks automatically, with built-in retry so a transient hiccup never loses work. Register symbols globally and search for usages across the codebase.

Semantic Code Search

Generate vector embeddings for project documents. Search by meaning, not keywords. Find "authentication logic" even when the code uses "auth", "verify", or "jwt" terminology. Integration with pgvector for enterprise-scale semantic search.

Intelligent Issue Management

Create an issue and it gets a clean identifier — `PROJ-42` — instantly, even with a dozen agents filing issues at once; a self-healing atomic counter prevents ID collisions before they happen. Before anything gets created, `get_similar_issues` checks whether the work already exists, so agents stop duplicating each other's tickets. Break a big issue into a parent with sub-tasks, link it to what it blocks or relates to, and track inline TODOs without ever leaving the issue.

Sprint Automation

Run a sprint's full lifecycle — create, start, complete — with velocity calculated automatically from what actually shipped, not what was estimated. When a sprint ends, `generate_release_notes` groups every change by type and `export_sprint_report` writes a clean markdown status update, so "what happened this sprint" is a generated artifact instead of a Friday afternoon spent scrolling through closed tickets. Incomplete issues get an explicit carry-over-or-drop policy instead of silently vanishing.

Agent Memory & Session Checkpoints

Every session starts with a full project brief and hands off cleanly to the next — zero ramp-up time for the agent that picks up the work. Checkpoint mid-task for safe interruption, `capture_learning` promotes a hard-won insight into the shared knowledge base, and `cleanup_stale_sessions` closes out idle sessions so the workload view stays honest about who's actually still working.

Knowledge Graph

Architecture decisions, conventions, gotchas, and reference material live in one searchable graph instead of scattered across READMEs nobody updates. Health monitoring flags entries going stale before they mislead the next agent that reads them, and semantic search surfaces exactly the relevant context the moment a new task starts — no more re-explaining the same gotcha to every agent that touches the code.

Spec & PRD Generation

Describe what you want in plain English and get a detailed spec or PRD back, tracked through its own lifecycle from draft to review to approved to implementing to completed — so "is this spec actually final" is never a guess. Architecture decisions get the same treatment through ADRs, tracked from proposed through decided to (eventually) superseded, so the reasoning behind a decision outlives whoever made it.

Bug Capture from Errors

A raw stack trace becomes a structured, filed bug report automatically — nobody has to manually transcribe an error into a ticket. A dedup check catches it if the same bug is hitting again, and blast radius simulation predicts which files and issues a fix will actually touch before anyone runs it.

Meta-Compiler — Build New Compiler Targets

Forge's visual Meta-Compiler (at `/forge/meta-compiler`) lets engineers add a brand-new language target to Aeion's Universal AST Engine without hand-writing the compiler. Drag a Language-Definition node, state variables, and AST node-cases onto a canvas; it emits a ready-to-register TypeScript `LanguageGenerator` class. The engine already ships **31 compilation targets** (PyTorch, GLSL, Solidity, IFC/BIM, eBPF, zk-SNARK, G-Code, MAVLink, FFmpeg…) — the Meta-Compiler makes that set open-ended: a compiler that writes compilers.

The 160+ MCP Actions — Grouped by Concern

Projects

list / get / create / update / delete · get_project_context · get_project_briefing · get_project_metrics · get_project_timeline

Issues

create / list / get / update / delete · batch_create / batch_update · get_issue_tree · search_issues · move_to_sprint · duplicate_issue · log_time · add_label / remove_label · assign_issue · get_my_issues · get_similar_issues · get_blocked_issues · get_unblocked_tasks · calculate_critical_path · clarify_issue · breakdown_issue · close_issue · create_subtasks · get_issue_history

Issue todos + links

add_todo / toggle_todo / remove_todo / set_todos · link_issues / unlink_issues / link_chain

Issue lifecycle

start_issue · submit_for_review · approve_issue · reject_issue · request_human_review · claim_issue · complete_issue · log_progress

Sprints

list / get / create · start_sprint · complete_sprint · suggest_sprint · add_sprint_todo / toggle_sprint_todo / remove_sprint_todo · export_sprint_report · generate_sprint_retro · generate_release_notes

Specs & PRDs

create_spec · get_spec · list_specs · update_spec · breakdown_spec · generate_spec

Decisions (ADR)

log_decision · list / get / update · accept_decision · supersede_decision · validate_adr_compliance

Knowledge

create / get / list / update / delete · search_knowledge · semantic_search_knowledge · get_relevant_knowledge · verify_knowledge · expire_knowledge · check_knowledge_freshness (code-drift + age + never-accessed) · check_knowledge_health · consolidate_memory (nightly pass, on-demand) · suggest_promotions · generate_embeddings

Diary

append_diary_entry · search_diary · promote_diary_to_knowledge

Agent memory + sessions

sync_agent_memory · list_agent_sessions · list_active_agents · get_agent_discoveries · capture_learning · checkpoint_session · break_session · complete_session · get_session_context · get_session_handoff · get_session_timeline · cleanup_stale_sessions · create_agent_briefing · get_agent_preferences / set_agent_preferences (standing reminders, briefing toggles, watched files, preferred test commands)

Code intelligence

index_codebase (glob) · reindex_changed_code (git-diff) · search_symbols · get_symbol_source · get_symbol_usage (transitive callers) · resolve_symbol_edges (call graph) · embed_symbols · semantic_code_search · register_symbol · register_file · commit_file_knowledge · get_file_info · search_file_registry · delete_file · claim_file_lock · release_file_lock · get_file_locks

Architecture audit

generate_architecture_diagram · audit_architecture · audit_codebase_smells · log_tech_debt · analyze_debt_hotspots · simulate_blast_radius

Bug capture + routing

create_bug_from_error · route_incoming_bug · run_issue_tests

Evidence + PR tracking

attach_commit (feeds the commits→files→issues graph) · reconcile_commits (git-log backfill) · link_pr · update_pr_status (with `autoClose` flag to close issue on PR merge)

Comments

add_comment · list_comments (with reply threading)

Checklists

create_checklist · update_checklist_item · list_checklists · get_checklist

Doc ingestion

ingest_docs · review_captures · find_conflicts · reingest_changed

Cross-project insight

find_cross_project_insights · generate_pulse · generate_retro

Workspaces (focus)

create / list / delete focus_workspace · get_workspace_context

Bounties + escalation

create_agent_bounty · request_human_review · submit_external_suggestion

Files

link_files · get_issue_files

Templates

list_templates · apply_template

Autonomous Development Workflows

Session Handoff

When an agent starts, it receives a full project brief including active sprint, velocity metrics, workload distribution, and all open issues. Zero ramp-up time.

Blast Radius Simulation

Before making cross-cutting changes, agents can simulate the blast radius—predicting which files, modules, and issues will be affected.

Codebase Smell Audit

Periodic automated analysis for technical debt, anti-patterns, and architectural drift. Issues auto-created for critical findings.

Critical Path Calculation

Identify which issues are blocking others and calculate the minimum viable release timeline.

PR Tracking

linkPR attaches GitHub/GitLab PRs to issues. updatePRStatus handles open/merged/closed transitions. merged+autoClose flag auto-closes the linked issue when PR merges.

Cross-Project Insights

Analyze patterns across multiple projects—shared technical debt, common blockers, knowledge that applies universally.

Built for Multi-Agent Orchestration

Aeion Forge is the coordination layer for the entire Aeion OS development workflow. When an AI agent receives a task:

Session Handoff — get_session_handoff returns full project context
Context Retrieval — semantic_search_knowledge finds relevant prior decisions
Dedup Check — get_similar_issues prevents redundant work
Work Execution — Create/update issues, generate specs, track time
Session Checkpoint — checkpoint_session preserves progress for next agent
Knowledge Capture — capture_learning promotes insights to shared knowledge base

The result: AI agents that operate like a cohesive team, not a collection of isolated tools.

160+ MCP Actions

Tree-sitter Code Parsing

Semantic Vector Search

Multi-Agent Memory Sync

Sprint Velocity Tracking

SOC 2 Type II ready (audit motion in flight)

Frequently Asked Questions

Linear / Jira are excellent for human teams. Forge is purpose-built for *AI agent teams* — agents read project context via the 160+ MCP actions, capture learnings to shared knowledge, hand off sessions to other agents with full context. Plus normal humans still use it via the same UI. It's project management that works for both.

Every agent session writes a structured handoff at completion — what was accomplished, what was tried that failed, the current mental model of the system, files for the next agent to read first, blockers. The next agent loads that handoff in one tool call and is up to speed without re-reading the whole conversation.

Code embeddings generated via the AI module's embedding providers, stored in pgvector. Search by meaning, not keywords — "the function that handles refund eligibility" surfaces the right code even if the actual function is named `evaluateClawback()`. Used by AI agents to find relevant code without random grep.

Wikis are write-then-read static documents. Forge knowledge is structured (category, tags, source files, expiration date), versioned (each edit creates a new version), validated (verify_knowledge against current code), and semantically searchable. Plus agents promote diary entries to knowledge after pattern-confirmation.

Yes — Forge works fine as a standalone PM tool. Issues, sprints, specs, decisions, knowledge, comments — all the standard PM functions. AI features are opt-in; you can disable agent-memory and AI generation while keeping the rest.

Cross-project tunnels — shared knowledge that applies to multiple projects (architecture patterns, language conventions). Pulse / retro generation summarizes activity across projects. Sprint reports + release notes auto-generated from issue activity. Designed for years-long projects, not just quarterly sprints.