Continue the conversation — chat opens pre-seeded with the current signal, caps, and movement.
Infrastructure subagent layer providing Fast Apply (morph-v3-fast/large, ~10,500 tok/s code merging, 98% accuracy, with an 'auto' complexity router), WarpGrep v2 (RL search subagent in an isolated context returning ~150-token precise spans, SEAL leaderboard traces), FlashCompact (context compression), Sandbox execution primitives, and — new in Jan 2026 — Glance, an RL browser-testing agent that posts AI PR-test videos into GitHub.
Target user: agent builders (Claude Code, Cursor, Cline, Codex, Aider, Windsurf, Amp, Antigravity, OpenCode) — not end developers. Picks-and-shovels positioning: Morph is infrastructure invoked by other agents, not a developer-facing tool.
Differentiator: vendor benchmark differentiation remains strong (Fast Apply ~10,500 tok/s, FlashCompact context compression); WarpGrep v2 publishes SWE-Bench Pro traces via Scale AI's standardized SEAL scaffolding showing a 2.1–3.7 point lift across frontier models — partial independent validation — and ships as an MCP server in any agent, a cross-agent distribution edge over Windsurf-locked competitors like Cognition's SWE-grep. AWS featured case study with Binance reference (5,000 developers) provides enterprise traction.
Adoption & Proof Points
- **AWS Customer Success case study** with Binance (5,000 developers) — 50–70% productivity gains reported
- **WarpGrep v2 SEAL leaderboard publication** (Scale AI standardized scaffolding) with 2.1–3.7 point SWE-Bench Pro lift across frontier models (partial independent validation); 15.6% cheaper, 28% faster via context isolation
- **NEW — Glance** (launched 2026-01-22): RL browser-testing agent + GitHub Action (morphllm/glance-preview@v1, latest v2 2026-02-04) that posts AI PR-test videos as PR comments; works with GitHub Actions / GitLab CI / any CI; free for OSS repos
- **Integration footprint:** Claude Code, Cursor (via MCP), Cline, Codex, Aider, OpenCode, Windsurf, Amp, Antigravity
- **First-party repos:** opencode-morph-plugin (56 stars, updated 2026-05-01), morph-claude-code-plugin (16 stars, 2026-04-21), examples (updated 2026-05-28), homebrew-tap (CLI)
- **SOC 2 Type II audit dated 2025-10-30**; AWS Marketplace listing
- **Goose-author endorsement on HN** (positive practitioner signal)
- **3-employee team, YC S23 batch; pre-seed/seed, no Series A; actively hiring 3 roles**
Risks & Limitations
- **Viability score 7 (binding constraint, held flat):** 3-employee pre-seed team, no Series A. (Prior eval's 'founder departure' downgrade trigger is confirmed FALSE — it referred to a different, crypto, company named Morph; Tejas Bhakta remains CEO.)
- **Cursor relationship is MCP-only:** direct integration moved to MCP — a competitive headwind
- **Vendor benchmark differentiation partly unverified against frontier-model native edits:** Fast Apply is fast and WarpGrep traces are partially validated via SEAL, but native LLM edit/search capability is closing
- **Glance adoption unproven:** the morph-glance-preview Action repo shows ~0 stars — brand-new, no third-party CI adoption proof yet
- **Picks-and-shovels positioning:** not a developer-facing tool — adoption depends on agent builders maintaining the integration
- **No SSO/SAML/RBAC evidence:** SOC 2 audit is dated but enterprise admin-control posture is unproven
- **Concentration risk:** Binance is the single named enterprise reference
Capabilities & Integration
**Autonomy (9, Semi-Autonomous):** Morph is an infrastructure subagent layer, not an end-to-end coding agent, but its subagents act autonomously within their narrow scope: Glance is an RL browser agent that decides what to test and drives browser interactions, and WarpGrep v2 runs multi-turn isolated-context search (up to 8 parallel calls/turn). Morph does not plan or self-correct across a full dev task — it is invoked by external agents.
**Integration (8, Basic Integration):** the new Glance GitHub Action (morphllm/glance-preview@v1) posts AI browser-test videos as PR comments, triggered by @morph/@glance comments, across GitHub Actions / GitLab CI / any CI — genuine PR/CI-workflow integration. Agent-mediated apply-layer integrations with Claude Code, Cursor (via MCP), Codex, Aider, Cline, OpenCode, Windsurf, Amp, Antigravity continue. No full PR review/merge or deep team-comms yet.
**Context (13, Repository-Deep):** WarpGrep v2 isolated-context RL search with 2.1–3.7 point SWE-Bench Pro lift across frontier models (SEAL traces). FlashCompact context compression. Cross-agent MCP server.
**Interface (13, Multi-Platform):** REST API (OpenAI-compatible), TypeScript SDK, MCP server, npm packages, AWS Marketplace, Claude Code plugin, OpenCode plugin, Homebrew tap (CLI), and the Glance GitHub Action.